docs: use american english spelling

[ChelseaKR]

Owner rule (2026-09-18): the portfolio uses American English everywhere, in code and in user-facing text.

Converted: 217 words in 77 files

behaviour→behavior (31), normalise/normalised/normalisation/unnormalised→normalize… (56), labelled/labelling→labeled/labeling (26), tokenise/retokenise…→tokenize… (21), analysed→analyzed in prose and Markdown output (12), centre→center (9), recognise/recognised/recognises→recognize… (8), serialise/serialiser/serialisable→serialize… (8), acknowledgement→acknowledgment (4), maximise/maximising→maximize… (5), grey→gray (2), plus canonicalisation, generalise, personalise, localised, fertilised, favour, honour, defence (2), tyres, and theatre.

The user-facing Markdown from sprout corpus diff now says "Not analyzed.". Test function names and private helpers are renamed. Nothing in the public sprout API (__all__) changes.

Kept with a reason: 57 occurrences in reviewed files

• 50 analysed: the field names analysed, not_analysed_reason, toxicity_analysed, and toxicity_not_analysed_reason in corpus_diff.py's pydantic models. They're the keys of sprout corpus diff --json, so they stay, and a one-line comment says so. Proposal: rename them the next time that JSON shape is versioned.
• 5 cancelled: GitHub Actions' conclusion value ("cancelled", backticked) in tests/test_ci_parity.py. The test's own name is converted.
• 1 categorisation and 1 harmonisation: verbatim EU AI Act terms ("biometric categorisation", "Annex I harmonisation legislation") in docs/audits/eu-ai-act-classification.md.

Not touched (third-party or corpus text)

• corpus/ and src/sprout/data/corpus/, the plant-care source text. Two hits, both "silvery-grey".
• eval/research/e3_external_suite_comparison/golden_set.json and eval/judge_probes.yaml, which quote that corpus verbatim. Ten hits.

Migrations and aliases

None were needed. No public name changed.

Verification (local)

Rebased onto current main, which already has the anyio 4.14.2 bump. The full make verify then passed in the pre-push gate ("verify: all gates green", 1314 tests). Before the rebase, the only failure was pip-audit flagging anyio 4.14.0, and main has since fixed that. gate-inventory regenerated docs/audits/gate-inventory.md, which picked up the behavior= stratum label.

Prepared with AI assistance; reviewed before submission.

[ChelseaKR]

Added 51a20f8 so the required tuning-scope gate passes. check-tuning-scope fingerprints the tunable modules by their AST, docstrings included, so the 15 docstring edits in retrieve.py, lexical.py, guards.py, confidence.py and providers/{base,bedrock,deterministic}.py read as tuning and would need a Tunes-Against citation that isn't true. Those docstring lines keep their original spelling (add them to the kept list: "docstrings in the tunable surface, because the tuning-scope gate reads any AST change there as tuning"). The comment-line edits in the same files stay, since comments are outside the AST. sprout check-tuning-scope --base origin/main passes locally, and so does the pre-push make verify. Then brought the branch up to date with main.