fix(sdk): address beta testing and migration gaps

[bokelley]

Summary

• forward production lifespan hooks and MCP session controls through the public in-process testing helpers
• normalize optional non-nullable None values to omission in version-pinned Pydantic extension models while preserving schema-nullable values
• add collision-safe semantic aliases and source-aware codemod mappings, with identity checks that refuse unsafe automatic rewrites

Why

SDK 8 beta adopter feedback exposed three production-parity and migration-safety gaps: test apps could not reproduce production lifespan/session configuration, versioned extension models rejected explicitly supplied optional None values despite advertising nullable Python annotations, and the v3-to-v4 codemod could substitute a different generated class when a bare public name collided.

Closes #1054.
Closes #1055.
Addresses the latest beta.5 follow-up in #911.

Validation

• three independent expert reviews: code, protocol/schema, and Python/Pydantic; no blockers found
• focused change suite: 333 passed; public API suite: 21 passed
• full pytest run: 6,740 passed, 40 skipped, 9 deselected, 1 xfailed; its two failures were the then-stale intentional API snapshot and an unrelated sub-microsecond wall-clock race, both of which passed on rerun after the snapshot update
• ruff check src/
• mypy src/adcp/
• mypy --strict tests/type_checks/
• pre-commit suite on the committed diff

[aao-secretariat]

Ladon verdict: Approve

Approve — beta-gap fix across codemod, version-scoped models, and testing helpers; no blocking or medium findings.

What I checked:

• Codemod (v3_to_v4.py) now identity-gates generated→public rewrites and fails closed on non-identical collisions; SKIP-hint collision path is covered by tests.
• Version-scoped extension models normalize explicit optional None to omission only when the schema forbids null; both omit and preserve branches are tested.
• In-process testing helpers forward MCP session controls (stateless_http / session_idle_timeout=1800.0 / max_active_sessions) plus lifespan hooks with matching production defaults, asserted against the production factories.

High-risk flag: true, but all four reasons are (modified) files with no medium-or-higher findings; per the change-kind guidance modifications with no medium concern are presumed safe. No (deleted) entries. gated_paths is false; no author-team gate.

Note (non-blocking, low): The additive public surface at aliases.py:2107 ships under fix(sdk): rather than feat:. This is a semver-signal nit — additive (non-breaking) surface only affects the version-bump magnitude, not adopter deserialization, so it does not meet the high breaking-change bar. Worth a maintainer glance but not blocking.

Decision table: no critical/high (row 1 no), gated_paths false (row 2 no), no deletions (row 3 no), no medium (rows 4/5/8 no), no prior escalate (row 6 no), no team gate (row 7 no) → row 9 approve.

[aao-secretariat]

Ladon verdict: Approve

Approve.

This is a subsequent push on a PR previously approved (prior head 36ccf4b, no findings). The current delta is a single test-only commit making two patch() call sites Python 3.10 compatible in tests/test_testing_decisioning.py — swapping string-path patch() for patch.object() on the importlib-resolved serve module, with wraps= preserving the real callable. Functionally equivalent; no production, wire-shape, or public-API surface touched.

Decision-table walk:

• Rows 1: no critical/high findings.
• Row 2: gated_paths is false — does not apply.
• Row 3: no (deleted) high-risk reasons — all high-risk entries are (modified).
• Row 4: no medium findings.
• Row 5: high_risk is true with (modified) entries, but no medium finding present — does not fire.
• Row 6: prior decision was approve, not escalate.
• Row 7: no no-auto-approve team match.
• Row 8: fewer than three medium findings (zero).
• Row 9: approve.

Note: the high-risk flag fires on the src/adcp/types/** and migrate files listed in the diff stats, but the reviewer's delta analysis confirms this incremental commit is test-only and clean, with no medium-or-higher concerns on any modified sensitive file. review_decision is REVIEW_REQUIRED but no hard path gate applies (gated_paths false), so a clean diff approves per row 9.