analyze.py: hardcoded English stopword/generic-basename lists

[hailcpy]

STOPWORDS and GENERIC_BASENAMES are hand-picked from fixtures. Real repos have their own generic files (handler.go, service.ts, base.py) not on the list → spurious merges. Non-English commit messages yield no useful tokens. Silent failure: wrong lists produce bad groupings with no error.

Fix: derive generic-file set from repo (files touched by many unrelated commits = generic), and/or make lists configurable.

[hailcpy]

The configurable half of this is done in the config refactor (PR incoming):

• stopwords / extra_stopwords and generic_basenames / extra_generic_basenames are now overridable via --config JSON (extra_* unions with defaults, bare key replaces).

The auto-derive-from-repo half (files touched by many unrelated commits = generic; language-agnostic tokenization for non-English messages) is the harder part and remains open. Suggest keeping this issue open for that, scoped down to the derivation work.