Root GC side tables and expand store-audit coverage

[andrewtdiz]

Summary

• make class/prototype/native-module side-table values precise barriered roots
• root class prototype-object and parent-closure side tables
• make symbol side-table values precise barriered roots
• resolve unaudited generated GC stores
• expand the GC store-site inventory scanner coverage

Stack note

This is a stacked draft after #2339, #2343, #2344, #2345, and #2348. Until those land, the GitHub diff against main includes earlier GC slices too.

Verification

• python3 scripts/gc_store_site_inventory.py --self-test
• bash tests/test_gc_store_site_inventory.sh
• cargo test -q -p perry-runtime gc::tests::runtime_roots --lib
• cargo test -q -p perry-runtime gc::tests::copying --lib
• cargo test -q -p perry-runtime gc::tests::cycle_state --lib
• cargo check -q -p perry-runtime -p perry-codegen -p perry-ffi
• cargo fmt --check
• git diff --check HEAD

[proggeramlug]

Heads-up: this one needs a rebase onto current main rather than a mechanical conflict-resolve, and I didn't want to guess inside the GC.

main's GC trigger path has evolved in parallel with this debt-pacing/cycle-state series, so the conflicts here are semantic, not textual — e.g. gc/mod.rs, arena/block.rs, and gc/policy.rs carry competing descriptions of the same path (legacy_synchronous on main vs. the debt-paced/bounded-assist behavior this PR introduces), plus new GcTriggerKind::Emergency and debt/pause telemetry fields that have to line up with whatever landed on main. Resolving those by hand risks introducing a GC correctness bug that the unit tests wouldn't reliably catch.

#2339 (pause/progress contract) and #2343 (resumable-cycle: ffi_code + budgeted_step_api tests) were mechanical and are merged/green. Could you rebase this part of the stack onto current main so the trigger-path behavior + telemetry reconcile against what's actually there now? I resolved & pushed the rest of your open PRs (~25 stream/buffer/util/url/assert/etc.) this round.