fix(getter): name C-family functions with nested declarators

[dekobon]

Fixes #1208

A C/C++/Objective-C function whose declared name sits under an extra
declarator layer resolved its FuncSpace::name to null. Two spellings
were affected:

int (*fp(int a, int b))(int c) { return 0; }        // a function returning a function pointer
void RUN_STATS_METHOD(allocate)(JNIEnv *env) { }    // a macro-obscured declarator (JNI shims)

get_func_space_name reached the declarator with first_occurrence, a
leftmost pre-order search. When the function node's own declarator
field is already a function_declarator — the return type's — the
search stops there, and its child(0) is a parenthesized_declarator
or an inner function_declarator, neither of which is an identifier
kind, so the arm fell through to None.

nargs has walked this chain correctly since #1200. That walk moves to
src/c_declarator.rs, and the four getters now read the name off the
same innermost declarator, so the arity and the name can no longer
describe two different nodes.

What changed

• src/c_declarator.rs (new) — innermost_declarator, moved out of
  src/metrics/nargs.rs with its node-types.json table doc intact,
  plus declarator_name, which is that node's declarator field. The
  three NArgs::params_owner impls and the Objective-C compute
  override point at the new path; no arity moved.
• The four getters read declarator_name::<Self>(node) and gate the
  result with matches! on their own identifier kinds. The
  FunctionDeclarator/2/3/4 enumerations are gone
  (grammar-dispatch.md §1) and the name comes off the declarator
  field rather than child(0) (§3). C++/mozcpp keep the OperatorCast
  early return: the shared walk deliberately cuts the chain at a
  conversion operator, so removing it would return None.
• A regression found during review — the fallback also runs on the
  C++ name forms, and template_function / template_method put
  their template_argument_list last, so a type argument spelling a
  function type pulled the chain off the name side.
  template <> void tspec<int (*)(int x, int y)>(int a) lost its name
  on this branch and had been misreading its arity since fix(nargs): C/C++ pointer/reference-return functions report nargs = 0 #1200.
  template_argument_list is now excluded alongside
  attribute_declaration.
• Search::first_occurrence loses its last production caller and
  trips dead_code on the lib target, so it sits behind a scoped
  #[allow] naming refactor(node): Search::first_occurrence has no production caller #1212. Both src/node.rs test docs that justified
  the pre-order contract by naming these four getters were rewritten —
  that claim is now false.

Measurements

Same method both sides (bca metrics --no-config over DeepSpeech +
pdf.js, 14,269 files, 232,198 spaces):

	nameless kind: function spaces
before	354
after	310

The issue's 354 is the nameless-function population, not this
defect's. This fixes 44 (46 gained a name, 2 lost one). The residual
310 is ERROR-recovery on input the grammars cannot parse:
schema_generated.h (115, a macro between a struct name and its base
clause), two SWIG .i files (56), gpu_device.cc (32). The clusters
the issue named as this defect — the JNI shims — are all fixed:
object_tracker_jni.cc (18), imageutils_jni.cc (5),
run_stats_jni.cc (4), plus 17 XLA files.

The 2 losses are both inside recovery subtrees, where no strategy has a
defensible answer; one had been reporting the callee of a misparsed if
statement as a function name. The boundary is documented on
innermost_declarator.

Snapshots

Two openfst union-weight.h snapshots move, where the issue predicted
none. Not a nameless space — a name→name change. tree-sitter-cpp cannot
parse friend bool operator== <>(const UnionWeight<W, O> &, …) and
folds the following comments and constructor into a
reference_declarator holding two sibling function_declarators; the
old name came from the leftmost, the new one from the last. That space's
function_args has read 1 since #1200 — the arity of
first_(W::NoWeight()), not of UnionWeight() — so the name and the
arity had been describing two different nodes and now describe one.
Submodule bumped in the same parent commit and pushed.

Tests

A table in src/c_declarator.rs over C / C++ / Mozcpp / Objective-C
asserting name and kind: function and span in one pass
(discharging grammar-dispatch.md §6): the three #1208 shapes, three
controls including the array-return int (*g(void))[4] that resolved
correctly before the fix, one row expecting no name, and eight C++
name forms. Mozcpp owns no file extension, so it reaches all of this
only through space_verbatim.

Reverting the getters fails 12 of 36 rows — exactly the three #1208
shapes × four languages. Removing the template_argument_list
exclusion fails exactly the two rows added for it.

a_parenthesised_declarator_still_loses_its_space_name, the #1200
bug-lock, is deleted: its fp fixture and plain control are rows in
the new table, in the module that now owns the behaviour.

Downstream note

Affected C-family functions re-key in .bca-baseline.toml from the
line-dependent <anon@L…> to their real qualified name, so a baseline
holding such an entry needs one refresh — after which the key is stable
across line drift like any other named function. bca functions stops
rendering these as a red error: line.

Follow-ups

• refactor(node): Search::first_occurrence has no production caller #1212 — Search::first_occurrence has no production caller.
• fix(nargs): macro-obscured declarators report the macro's arity, not the function's #1213 — the macro-declarator arity question. nargs reads
  RUN_STATS_METHOD(allocate)(env, clazz) as one argument; this makes
  the name agree with it, so the two surfaces are consistent. Whether
  they are consistent about the right node is open.

make pre-commit → BCA_GATE: pass.

[codecov]

Codecov Report

❌ Patch coverage is 97.69231% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 98.31%. Comparing base (859be0e) to head (dfb2352).

Files with missing lines	Patch %	Lines
src/c_declarator.rs	96.90%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1214      +/-   ##
==========================================
+ Coverage   98.30%   98.31%   +0.01%     
==========================================
  Files         276      277       +1     
  Lines       71858    71904      +46     
  Branches    71428    71474      +46     
==========================================
+ Hits        70643    70696      +53     
+ Misses        802      796       -6     
+ Partials      413      412       -1     

Flag	Coverage Δ
rust	98.30% <97.69%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/getter/c.rs	85.71% <100.00%> (+2.85%)	⬆️
src/getter/cpp.rs	100.00% <100.00%> (+4.44%)	⬆️
src/getter/mozcpp.rs	94.73% <100.00%> (+3.62%)	⬆️
src/getter/objc.rs	84.84% <100.00%> (-0.16%)	⬇️
src/metrics/nargs.rs	99.50% <100.00%> (+0.08%)	⬆️
src/node.rs	98.42% <ø> (ø)
src/c_declarator.rs	96.90% <96.90%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.