verify: warn on bare fmt/fmt2 at non-tail position (ILO-T032)

[danieljohnmorris]

Summary

fmt and fmt2 are pure-functional sprintf builders. A bare fmt "..." v statement evaluates and discards the resulting text on every engine - nothing reaches stdout. Verifier now flags this with ILO-T032 and a hint pointing at the two idiomatic fix shapes (prnt fmt ... / name = fmt ...).

This is a diagnostic-only change. Runtime behaviour across tree, VM, and Cranelift is unchanged.

Persona-framing correction

The persona report that prompted this (ilo_assessment_feedback.md line 2138, quant-trader rerun3) framed the silent-discard as Cranelift-specific - "on 0.11.2 from the Cranelift default path, bare fmt is silent." That framing is wrong. I reproduced the same shape (go>t;fmt "discarded={}" 42;prnt "after") on all three engines and observed identical behaviour: the fmt result is discarded, only prnt "after" reaches stdout. This is correct functional semantics (bare expression statements don't perform I/O), consistent with the existing precedent that bare +=xs v and bare mset m k v don't mutate either. This PR doesn't change the behaviour; it adds the diagnostic the persona explicitly asked for.

Repro before/after

report v:n>n;fmt "v={}" v;prnt "done";v

Before: runs silently to completion, stdout = done\n7. The fmt "v={}" v evaluates and is thrown away. The agent spends ~3 minutes diagnosing why their report function emits nothing.

After: verifier emits ILO-T032 bare 'fmt' result is discarded with hint did you mean prnt fmt ... to print, or name = fmt ... to capture?. Program still runs (warning, not error), but the agent now sees the problem immediately.

What's in the diff (per commit)

• verify: warn on bare fmt/fmt2 at non-tail position (ILO-T032) - the verifier check itself, the diagnostic registry entry with examples, and six unit tests covering bare fmt warns, bare fmt2 warns, tail-position no warn, bound-to-name no warn, nested-in-prnt no warn, multiple bare fmts each warn independently.
• example: bare-fmt-warns.ilo pins the two idiomatic shapes - new examples/bare-fmt-warns.ilo demonstrating prnt fmt ..., name=fmt ...;prnt name, and tail-position fmt, exercised across tree + VM via tests/examples_engines.rs.
• docs: note that fmt is pure-functional sprintf, not print - SPEC.md, ai.txt, and skills/ilo/SKILL.md all gain an explicit "fmt does not print" note next to the existing fmt entry, with a pointer to ILO-T032 and the fix shapes. (Site builtins.md updated in a separate commit/PR in the site repo.)

Test plan

• cargo test --release --features cranelift --lib verify::tests::bare_fmt verify::tests::fmt_in_tail verify::tests::fmt_bound verify::tests::fmt_inside verify::tests::multiple_bare_fmts - 6/6 pass
• cargo test --release --features cranelift --test examples_engines - examples_engines passes (new example included)
• cargo test --release --features cranelift --lib vm::compile_cranelift::tests::aot - all 7 AOT tests pass against default target dir
• cargo fmt --check - clean
• cargo clippy --release --features cranelift --all-targets -- -D warnings - clean
• Manual: ran go>t;fmt "discarded={}" 42;prnt "after" on all three engines (--run-tree, --run-vm, --run-cranelift) - confirmed warning surfaces at CLI and program still runs (exit 0).

Follow-ups

• +=xs v and mset m k v bare-discard family (assessment doc line 2598) is a clear next PR, same shape: warn when non-tail Stmt::Expr is a discarded result-producing builtin call. Deferred per scope decision so this PR stays minimal and easy to revert if needed.
• Loop-body discards (where fmt is the only statement in a foreach/while body) also discard silently but are not warned by this PR. Same follow-up could widen the check to cover that case.

[codecov]

Codecov Report

❌ Patch coverage is 98.82353% with 1 line in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/verify.rs	98.82%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!