fix(examples): ys-lower-report calls the public compile-form

[mparrett]

examples/aot/yamlstar/ys-lower-report.lg has been unable to run since
compile-form* became private. Every invocation dies at load:

error: compiling let body
caused by: Can't resolve ir.passes.pipeline/compile-form* in this context
  --> examples/aot/yamlstar/ys-lower-report.lg:45:32

It never reports on a single form, so the per-fn lower/fallback report described
in examples/aot/README.md is currently unavailable.

Why the public wrapper, beyond it resolving

compile-form normalises a :multi-fn-template result to :status :lowered.
Its own comment gives the reason: public callers key off :status, and without
the normalisation "a form that lowered via lambda-lift reads as unlowered".
ys-lower-report is one of those callers — it branches on :status and prints
anything other than :lowered as a fallback. So the wrapper is the correct
entry point here, not just the reachable one.

Verification

Ran the tool over five in-repo namespaces, which exercises the whole of its
contract:

File	Result
examples/aot/native-entry/fib.lg	ok=2 fallback=0 threw=0
examples/aot/cross-package/src/lib.lg	ok=2 fallback=0 threw=0
test/native-entry/closure_capture.lg	ok=2 fallback=0 threw=0
test/native-entry/loop_carried_closure.lg	ok=3 fallback=0 threw=0
test/native-entry/conditional_closure.lg	ok=4 fallback=0 threw=0

fib.lg's ok=2 agrees with the 2 fns lowered that lg-compile reports for
the same file.

The three closure fixtures are the lambda-lift shapes, and they only show that
the fixed tool reports them lowered. There is no before-and-after to compare,
because nothing ran before.

Scope

Two lines: the call site and the header comment that describes it. Other
references to compile-form* in the tree describe the private function itself
and are left alone.

Found while using the tool to survey a demo's lowerability against tip, which
also produced #796 and the comments on #551, #562, #660 and #358.

[nooga]

Review summary

Verified end to end:

• Reverted the fix locally and reran the tool — reproduces the exact reported error (Can't resolve ir.passes.pipeline/compile-form* in this context at the same line).
• Restored the fix and ran it against all five files listed in the PR body (fib.lg, lib.lg, and the three closure fixtures) — output matches the PR's table exactly, including ok=2/ok=3/ok=4 and fallback=0 threw=0 across the board.
• Confirmed the compile-form vs compile-form* distinction actually matters here, not just resolves: read compile-form's normalization in pkg/rt/core/ir/passes/pipeline.lg (a :multi-fn-template result gets :status :lowered attached, since the private path keys off :kind instead) and confirmed the three closure fixtures are exactly the lambda-lift shape that would read as unlowered without it.
• Checked scope: grepped for remaining compile-form* references in the tree — all are the definition itself, doc/comment mentions, or internal test files that intentionally exercise the private function directly, not other broken external callers.
• go build ./..., go vet ./... clean.

Small, precise, well-verified fix. LGTM — approving.

[nooga]

Verified: reproduced the original error, confirmed the fix resolves it and matches the PR's reported output across all five test files, and confirmed compile-form's :multi-fn-template normalization is the correct (not just reachable) entry point here.

[nnunley]

Approved and green at cb90b439a05e2d3ee04ac1ae3709babe8d31747f, but GitHub currently reports this branch as behind main (now 3aa9f4ea86d41dd55ced66fb3ca800c4bc66c355 after #800).

@mparrett, please update/rebase the contributor-owned branch and let CI rerun. I will not push to it. Once it is current-base and fully green, it is next eligible to land.

[mparrett]

Rebased onto 3aa9f4ea and force-pushed; head is now 1710dacc. Clean rebase, diff unchanged — the same two lines in examples/aot/yamlstar/ys-lower-report.lg.

Re-verified on the new base. The pre-fix script still fails at the same line, and the fixed version reproduces the PR body's table: fib.lg ok=2, lib.lg ok=2, closure_capture.lg ok=2, loop_carried_closure.lg ok=3, conditional_closure.lg ok=4, fallback=0 threw=0 throughout.

CI is green on the new head: all 15 checks pass.

[nnunley]

Thanks for updating this to 1710daccaea872d062c8f722538bba2e8d479ca2; its fresh checks are all green. Main advanced once more when the previously blocked #764 cleared and merged as 638b4a6aa1ab0aaadcfa42d29458e02bcbcd7bdd, so the PR is again one base step behind our exact-current-base merge rule.

@mparrett, please update the contributor-owned branch once more and let CI rerun. I will not push it. I’ll merge after confirming the new head/base, approval, mergeability, and every required check.

[mparrett]

Updated again: head is now 9ee86d1f, a single commit on 638b4a6a.

The branch had picked up a merge commit from the Update-branch button (acfe213b, "Merge branch 'main' into fix/ys-lower-report-public-api"). I replaced it with a rebase, so the branch is one commit on current main rather than a merge. main requires linear history here, and it keeps the head/base check unambiguous.

Re-verified on the new base, since #764 changes small-map ordering and the IR pipeline reads those maps: the five files still report ok=2, ok=2, ok=2, ok=3, ok=4, with fallback=0 threw=0.

All 15 checks green.