fix: prefix ?? accepts call expression as value side

[danieljohnmorris]

Summary

Prefix ?? (nil-coalesce) was parsing its value side with parse_operand, which only consumes an atom. So the most common nil-coalesce shape in agent-written ilo, ??mget m "k" 0, mis-parsed: mget was taken as the value atom and m "k" 0 as the default expression, which then failed because m is a Map, not a function.

Workarounds existed (??(mget m "k") 0, bind-first), but every site cost extra tokens and agent attention. Manifesto framing: this is a parser-precedence tax paid on every map lookup with a fallback, and it can be removed for free with the arity-cap pattern the parser already uses in four other places.

Repro before/after

main>n;m=mset mmap "k" 42;v=??mget m "k" 0;v

Before (main):

{"code":"ILO-T004","message":"undefined variable 'mget'", ...}
{"code":"ILO-T005","message":"'m' is a M _ n, not a function (called with 2 args)", ...}

After (this branch): 42

What's in the diff

Three commits, one logical change each:

1. parser: prefix ?? accepts call expression as value side — swaps the value-side parser in both prefix-?? arms (parse_expr_inner and parse_operand) from parse_operand() to parse_call_arg(false, None). parse_call_arg is a strict superset: it handles every operand case and additionally expands a known-arity ident into a call consuming exactly its declared arity. This mirrors the existing arity-cap behaviour in no_whitespace_call (list literals) and parse_call_arg's own nested-call block. Identifier-only ??x default, paren ??(expr) default, infix c??d, and chained f a ?? g b ?? d all keep working because parse_call_arg falls through to parse_operand for non-known-arity idents.

2. test: prefix ?? with call-expression value across all engines — extends tests/regression_prefix_nil_coalesce.rs with four new cases (hit, miss, two-call chain, paren workaround still works) running on tree / VM / Cranelift.

3. example: prefix ?? with map lookup and default — examples/qq-call-default.ilo, pinned by the engine harness. Three -- run: entries (hit, miss, chain) act as both regression coverage and an in-context demo for agents.

Test plan

• Repro fixed on tree, VM, Cranelift (verified manually before commit)
• cargo test --release --features cranelift full suite green (3071 + 340 + others = 0 failures)
• cargo fmt --check clean
• cargo clippy --release --features cranelift --all-targets -- -D warnings clean
• Targeted regression file passes 3/3 engine variants
• Examples harness picks up the new example file and passes
• Existing workarounds (parens, bind-first) confirmed still working

Follow-ups

None. Adjacent observation: ??-x default (prefix-minus value side) is broken on main too, but that's a separate operand-handling issue and out of scope for this fix.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

📢 Thoughts on this report? Let us know!