add ILO-P102 hint for top-level bindings without main wrapper

[danieljohnmorris]

Summary

K-means and linear-regression personas both got stuck on the same misparse this week: a top-level imperative chain written without a `main>_;` wrapper. The parser either died on the bare `=` (a bare ILO-P003 with no actionable hint) or, when a prior `name>type;body` decl was in scope, slurped the whole chain into that fn's body and emitted a wall of misleading ILO-T005 cascades anchored on the wrong line. K-means hit the 600s watchdog because of the cascade. Pending #5aj.

This adds ILO-P102, which catches the shape at parse time and points at the `main>_;` fix.

Repro before/after

```
gen-pts>L(L n);[[2.0 3.0][8.0 8.0][0.5 0.5]]
iter cs:L(L n) pts:L(L n)>L(L n);cs
pts=gen-pts;cs0=[[4.8 4.9][6.2 7.1][1.0 1.0]];cs1=iter cs0 pts;cs2=iter cs1 pts;prnt cs2
```

Before (2 diagnostics, anchored on the wrong function):

```
ILO-T005: 'cs' is a L L n, not a function (called with 1 args) (in function 'iter')
ILO-T039: 'gen-pts' is a 0-arg function used as a value reference, not a call (in function 'iter')
```

After (1 diagnostic, anchored on the orphan line):

```
ILO-P102: top-level `pts=...` binding outside any function declaration
hint: ilo programs need a function header. Wrap imperative chains in `main>;`
(e.g. `main>;pts=...;...;prnt result`) or give this binding a proper
signature like `pts>n;`
```

What's in the diff

• `add ILO-P102 for top-level bindings without main wrapper` - two parser changes:
  • `can_start_operand` refuses to cross an un-indented newline into an `Ident=` binding shape, so a prior fn's body can't slurp the orphan chain as a call argument. This kills the T005 cascade at the source.
  • `parse_decl` recognises a bare `Ident=` at the top level (after the existing reserved-keyword / builtin / alias guards) and emits ILO-P102 with the `main>_;` wrapper hint.
  • Registry entry for ILO-P102 with the long-form `--explain` text.
• `add ILO-P102 regression test and example` - cross-engine regression covering both repro shapes (bare top-level chain + chain slurped into a prior fn body), plus a workaround test (`main>_;` wrapper runs cleanly), a negative test (normal fn decl unaffected), and a guard test (`map=...` keeps its existing ILO-P011 builtin-shadow hint). New `examples/top-level-chain-hint.ilo` shows the corrected shape as agent-facing learning material.
• `doc sync for ILO-P102` - CHANGELOG entry under Unreleased, plus a row in the SPEC cross-language gotchas table.

Parser-only change; output is identical across VM and JIT.

Test plan

• `cargo test --release --features cranelift` clean
• `cargo clippy --release --features cranelift --all-targets -- -D warnings` clean
• `cargo fmt --check` clean
• New `tests/regression_top_level_chain_hint.rs` passes on `--vm` and `--jit` (5 sub-checks each, plus a parser-only `ilo check` smoke)
• `tests/examples_engines.rs` keeps passing (1004 example-engine combos) - confirms un-indented body continuations like `examples/apps/config-shaper.ilo` still parse cleanly

Follow-ups

• None for this PR. The cascade-suppression infra in `parse_program` already covers P102 via `sync_to_decl_boundary` walking to EOF for orphan chains, so a single P102 covers any length of orphan run.

[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!