feat(slc): accept end=-1 as 'to end' sugar (#15)

[danieljohnmorris]

Summary

Pending entry P2 #15: slc s n -1 doesn't mean "to end". Today ilo's slc reads negative end indices Python-style (slc xs 0 -1 drops the last element), but agents trained on Python/JS keep reaching for -1 to mean "to end of list/text", and the friction shows up in persona transcripts.

This PR special-cases one narrow shape: when start >= 0 AND end == -1, end is read as len xs. Every other shape (negative start, or end <= -2) keeps the existing Python-style relative-offset semantics, so slc xs -3 -1 is still the penultimate window and slc xs 0 -2 still drops the last two.

Manifesto framing: matches the agent's first guess, zero extra tokens at the call site, narrow enough that the existing Python-style negative-offset path stays intact for the cases that already depend on it. "Drop last" continues to live on take -1 xs, which I now point at from the docs.

Repro before/after

Before:

$ ilo 'f xs:L n>L n;slc xs 0 -1' '[10,20,30,40,50]'
[10, 20, 30, 40]

After:

$ ilo 'f xs:L n>L n;slc xs 0 -1' '[10,20,30,40,50]'
[10, 20, 30, 40, 50]
$ ilo 'f xs:L n>L n;take -1 xs' '[10,20,30,40,50]'   # new shape for 'drop last'
[10, 20, 30, 40]
$ ilo 'f xs:L n>L n;slc xs -3 -1' '[10,20,30,40,50]' # negative start kept Python-style
[30, 40]
$ ilo 'f xs:L n>L n;slc xs 0 -2' '[10,20,30,40,50]'  # end<-1 kept Python-style
[10, 20, 30]

What's in the diff (per commit)

1. slc: accept end=-1 as 'to end' sugar when start is non-negative - new resolve_slc_end helper in src/builtins.rs; tree-walker (src/interpreter/mod.rs), VM OP_SLC handler, and Cranelift JIT jit_slc all route through it.
2. tests: pin slc end=-1 sugar and adjacent Python-style shapes - cross-engine regression suite updated in tests/regression_neg_index_slice.rs. The existing slc xs 0 -1 / slc "hello" 0 -1 / quant-trader fencepost tests now assert the sugar; new pins for slc xs 2 -1 (mid-start sugar), slc xs -1 -1 (negative start keeps Python), slc xs 0 -2 (end<-1 keeps Python), and take -1 xs as the canonical drop-last shape. 34/34 pass on every available engine.
3. examples: slc-to-end demo + refresh negative-indices for the new sugar - examples/slc-to-end.ilo for the engines harness; examples/negative-indices.ilo switches its drop-last and trim-edges demos onto take -1 / explicit (- (len s) 1) and adds a penultimate example that exercises the negative-start path.
4. docs: document slc end=-1 'to end' sugar in SPEC, ai.txt, and skill - SPEC.md slc row, ai.txt slc line, skills/ilo/ilo-builtins-core.md list section. Companion PR on the site repo (already merged) updates docs/builtins/collections.md.

Behaviour change worth flagging

This is a behaviour change for one specific shape (slc xs 0 -1 was Python-style "drop last" before this PR). I checked every test that hit that shape; the only callers were three regression tests (added in PR #183) and the negative-indices.ilo example, all of which I updated to the new contract or moved onto take -1. The quant-trader-fencepost flagship example in the original PR moves to take -1 eq, which is the shape I'd reach for now.

If we want to add a deprecation hint for slc xs <non-neg> -1 users who are on the old contract, we could surface a one-time advisory the first time they hit the shape - happy to layer that as a follow-up, but the verifier doesn't track call-site provenance for builtins right now.

Test plan

• cargo test --release --features cranelift - full suite green (200 result groups, 0 failures)
• cargo fmt --check - clean
• cargo clippy --release --features cranelift --all-targets -- -D warnings - clean
• examples/slc-to-end.ilo runs on every engine via tests/examples_engines.rs
• examples/negative-indices.ilo updated assertions run on every engine

Follow-ups

• Consider a deprecation hint or comparison example for the old slc xs 0 -1 drop-last shape if persona transcripts show agents still reaching for it (would land separately once we have signal).

[codecov]

Codecov Report

❌ Patch coverage is 88.23529% with 2 lines in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/interpreter/mod.rs	71.42%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!