feat(trace): tape format v2 — scope-qualified locals via S records (#539 v2)

[InauguralPhysicist]

The v2 half of #539 (v3 — the src/eigsdap.c DAP server with stepBack/reverseContinue in VS Code — remains open on that issue, and now has the scope-qualified substrate it was blocked on).

The problem

A-records carry the name only, so same-named bindings from different scopes merged into one stream: --step (and any tape consumer) could not tell a function-local i from the top-level one, and trajectory labels fed on the interleaved garbage.

The format change (TRACE_FORMAT_VERSION 1→2)

A new scope-transition record qualifies the A records that follow:

S <fn> <depth> <serial>

<serial> is a per-thread monotonically increasing frame-instance id (CallFrame.call_serial, stamped at all four frame-push sites) — so two invocations of the same function never merge either. Emission mirrors the L-record dedup discipline: an S is written only when the frame owning the next assignment differs from the last S, so the byte cost lands at call boundaries that actually assign.

• Measured (call-heavy worst case, 20k calls, n=5): +43% tape bytes, ~8% traced runtime. A straight-line module tape carries exactly one S; untraced execution is untouched (the hook stays off). Stripping S records reproduces the v1 tape byte-for-byte — the change is purely additive.
• Replay skips S like A — byte-exact replay determinism unchanged (verified same-seed).
• v1 tapes refuse on both replay and --step per the Design: trace-tape format versioning — the tape is becoming a persisted artifact #411 version-and-reject rule (no compat promise, ever).

The stepper

--step folds S records into a reconstructed call chain (returning into a live serial pops to it; a new serial pops same-or-deeper frames and pushes with parent linkage):

• p shows the live chain's bindings innermost-first with {in fn} notes and shadowing — dead frames' locals no longer appear, and a shadowed outer i stays hidden.
• p name / t name resolve innermost-first, so trajectory labels are per-invocation ((i in work, frame #3)).

Pins & docs

• tests/test_step.sh grows six scope checks (22 total): S records present, distinct serials per invocation, module stream unpolluted, frame-local shadowing, per-frame trajectory resolution, shadowed-outer hidden.
• docs/TRACE.md: the S record row, the A record's new semantics, and the format version history.

Validation

Release suite 2737/2737; ASan+UBSan detect_leaks=1 2737/2737, zero LeakSanitizer reports (the run caught a t->scopes teardown leak in the stepper during development — fixed at all three teardown sites). Rebased on b5db2d6 (#547).

🤖 Generated with Claude Code

https://claude.ai/code/session_01GdDAVfZHgoQ2UNP63FYUXm