feat(vm): populate debug.getinfo name/namewhat from the call site

[davydog187]

Goal

Make debug.getinfo(level, "n") return a non-nil name and a matching namewhat, so the issue repro passes:

function F(a) return a end
F(1)
assert(debug.getinfo(1, "n").name == 'F')

Success criteria

• debug.getinfo(1, "n").name == "F" for function F() ... end; F()
• namewhat reports global / local / upvalue / field / method matching the call site, and "" when no name is known
• Regression unit test pins the issue repro
• constructs.lua skip narrows past the debug.getinfo block (was 225..313, now 232..313)
• mix test and the lua53 suite stay green

Changes

The caller's call instruction already carries a compile-time name hint ({:global, "F"}, {:local, "x"}, ...) — the same hint that powers attempt to call a nil value (global 'foo') errors. Two gaps kept it from reaching debug.getinfo:

• The interpreter's :call handler pushed a call_stack frame for :lua_closure callees but routed :compiled_closure callees straight into Dispatcher.execute without recording a frame. A top-level function F() ... end; F() therefore ran with an empty call_stack, so getinfo had no name to report.
• call_stack frames stored only the textual name, not the hint tag needed to classify namewhat.

This PR:

• Records a call_stack frame for compiled-closure calls in the interpreter (mirroring the existing :lua_closure push), so the running function's name is always available.
• Threads the hint tag through to a new namewhat field on each frame, derived by hint_namewhat/1 in Lua.VM.Executor.
• Reads the running function's frame (the head of call_stack when a native callback executes) in Lua.VM.Stdlib.Debug to surface name/namewhat.

No prototype/codegen change was needed: the call-site hint already lives on the instruction stream, which is the faithful PUC-Lua getfuncname model.

Verification

mix format
mix compile --warnings-as-errors   # clean
mix test                           # 2040 passed, 22 skipped (was 2037)
mix test test/lua53_suite_test.exs --only lua53   # 14 passed, 15 skipped

Out of scope

• The full PUC-Lua getfuncname runtime instruction walk for every form. We reuse the existing compile-time hint, which covers global / local / upvalue / field / method call sites.
• Naming of functions reached through a call site that carries no hint (call through a temporary, (t[expr])(), immediately-invoked anonymous closures) and functions reached by a tail call: these keep name == nil / namewhat == "", matching PUC-Lua's unknown fallback.
• name/namewhat for the entry chunk and for native (C) functions inspected by value (debug.getinfo(fn, "n")).

Closes #279

[davydog187]

Review: feat(vm): populate debug.getinfo name/namewhat from the call site

Reviewed for correctness, tests, conventions, and simplification. Verdict: LGTM, ship it. No blockers.

What I verified

• Regression tests genuinely pin the bug. Ran the three new debug_test.exs cases against the pre-fix source (main's debug.ex + executor.ex): 3 of them fail (name == nil / all-nil namewhat), and they pass on the branch. The "no hint" case correctly passes both ways since it asserts nil / "".
• Level-to-frame indexing is correct. Name uses call_stack[level-1] while position uses call_stack[level-2] — these legitimately differ because a frame stores the callee's name (from the call instruction's name_hint) but the caller's line/source. Walked chunk→A→B→getinfo by hand; level 1/2/3 all resolve to the right name, namewhat, and currentline.
• The compiled-closure frame push is sound. Dispatcher.do_execute_top does not push a frame or bump call_depth on entry — it relies on callers to do so, exactly as the dispatcher's own internal call sites (push → execute → pop) do. The interpreter's :call handler previously skipped this for :compiled_closure, so the fix also (correctly) makes the interpreter→dispatcher hop respect check_call_depth!. Pop ordering mirrors the existing sites.
• Full validation green: mix format --check-formatted, mix compile --force --warnings-as-errors (clean), mix test (2040 passed, 22 skipped), mix test --only lua53 (14 passed, 15 skipped).
• Skip narrowing is minimal and real. constructs.lua 225..313 → 232..313 un-skips the debug.getinfo(1,"n").name == 'F' block (lines 225-231), which now runs as part of the suite and passes. Reason string dropped the stale debug.getinfo mention; issue field stays nil.

Conventions: all clean

• Title/feat scope is vm (allowed subsystem), not a plan id. chore(A44) commits touch only the plan file, which is the documented exception.
• No plan-id references in any source/test the PR adds.
• No Co-Authored-By trailers.
• Body has Closes #279.

Minor (non-blocking)

• hint_namewhat/1 has 2-tuple clauses {:field, _} and {:method, _}, but the compiler's name_hint/2 only ever emits the 3-tuple forms {:field, name, recv} / {:method, name, recv}. The 2-tuple clauses are dead, though they harmlessly mirror the defensive style in format_target_hint/1.
• The value-based path (debug.getinfo(somefunc, "n")) still omits namewhat entirely, so it reads as nil in Lua rather than PUC-Lua's "". Pre-existing and explicitly out of scope here.