fix(stdlib): return error instead of panicking on windows size 0 🪟

[timfennis]

What

windows(seq, 0) panicked the VM instead of returning an error:

$ echo 'windows([1,2,3], 0)' | ndc
thread 'main' panicked at ndc_stdlib/src/sequence.rs:716:18:
window size must be non-zero

The stdlib windows function took length: i64, cast it to usize, and passed it straight to slice::windows(length) — which panics when the size is 0. Negative input was also wrong: it wrapped to a huge usize and silently returned [].

The sibling chunks function already handles both cases correctly (a length == 0 guard plus a usize parameter so negatives are rejected at argument conversion). This change brings windows in line.

Changes

• ndc_stdlib/src/sequence.rs: windows now takes length: usize and returns an error when length == 0. Negatives are now rejected at arg conversion (arg 1: expected a non-negative integer), matching chunks.
• tests/functional/programs/900_bugs/bug0024_windows_zero_size.ndc: regression test asserting the graceful window size must be non-zero error.

Context

Found while hunting for a way to panic the pipeline (lexer → parser → analyser → compiler → VM). The existing token-stream fuzzer never reaches this code because it only generates the identifiers a,b,c,x,f,g, so it never calls any stdlib function by name — the whole stdlib surface is outside its reach.

Behaviour unchanged on the happy path: windows([1,2,3,4], 2) → [[1,2],[2,3],[3,4]], and a too-short sequence still yields [].

🤖