v0.0.345: sys.settrace / sys.setprofile (F7)

[tamnd]

Summary

Closes group F7 from the v0.1.0 audit (notes/Spec/1500/1530_goipy_v01_gap_audit.md). Spec at notes/Spec/1500/1539_goipy_v0045_settrace_setprofile.md.

pdb, coverage.py, and third-party profilers (yappi, line_profiler) drive their hooks through sys.settrace/sys.setprofile. Goipy had stub no-ops in threading and pdb and nothing on sys itself; tracers silently never fired.

• Interp.traceFn / profileFn plus an inTrace re-entry guard so the tracer's own dispatch doesn't recurse.
• runFrame fires 'call' / 'return' / 'exception' events; the global trace's 'call' return value becomes the per-frame LocalTrace.
• Frame.LocalTrace and Frame.LastLine; LastLine is seeded to Code.FirstLineNo on frame entry so the RESUME-on-def-line opcode does not fire a stray line event before the first body line — matching CPython 3.11+.
• Dispatch hot-loop guard: nil LocalTrace is the common path → near-zero overhead. A local trace returning None disables further line events on that frame; a callable replaces the local trace.
• callFunctionFast frame-pool reset clears LocalTrace + LastLine.
• Frame f_trace getter/setter wired via LocalTrace.
• sys.settrace / gettrace / setprofile / getprofile.
• Fixture 345_settrace_setprofile.py — gettrace round-trip, event sequence for f(), per-frame replacement, profile (no line events), disabling mid-execution.

Test plan

• go test ./... -timeout 600s — green
• Fixture matches CPython 3.14 byte-for-byte
• No regression on existing fixtures (especially generator + exception tests)

Out of scope

• 'opcode' event (f_trace_opcodes stays a no-op).
• sys.monitoring (PEP 669) — separate audit entry.
• frame.f_lineno = N jump during pdb.