fix(affected): honor custom filter glob semantics

[Jseca666]

codegraph affected --filter 'tests/**/*.py' misses tests directly inside tests/, while --filter 'tests/*.py' can also select other/tests/... or a changed .py.bak file. This change uses the existing picomatch dependency to match glob patterns correctly.

Filters containing / match the whole project-relative path, and **/ matches zero or more directories. Filename-only filters such as *_test.go continue to match at any depth, and explicit custom filtering still includes dotfiles.

This PR now focuses on the remaining custom-filter fix from #1273. The default test-file recognition is already covered by #1803; its shared isTestPath classifier and existing Go/Python/JVM regression tests are retained. The branch is synchronized with main and preserves the original contribution history.

Validation

• npm run build passed on Windows with Node 24.14.1.
• 50 tests passed across cli-affected-filter, cli-affected-paths, cli-affected-test-conventions, is-test-file, and graph.
• The new CLI regression cases reproduced missing root-level matches and incorrect directory matches against the upstream implementation before the fix.
• git diff --check passed.
• npm audit reports 10 existing dependency advisories (1 critical, 4 high, 5 moderate). Dependency manifests and lockfiles match upstream; this PR adds or updates no dependencies.

[maxmilian]

Drive-by note from #1507, in case it helps this get looked at: this PR also fixes a bug nobody has connected to it.

#1507 (opened today) reports codegraph affected missing math_test.go for a Go file. That is the same inline defaultTestPatterns block you delete here — Go's underscore suffix matches none of the six JS/TS patterns, and the shared isTestFile you import matches it on /[._-](test|tests|spec|specs)\.[a-z0-9]+$/. So this PR closes at least two issues, not one.

I ran both predicates against the same inputs at upstream/main (c65d56c) to be sure rather than reading the regexes:

path	inline (today)	shared (this PR)
math_test.go	false	true
tests/test_math_utils.py	false	true
FooTest.kt	false	true
math.test.ts	true	true
math.go	false	false

The JVM/Swift CamelCase shapes (FooTest.kt, BarTests.swift, QuxTestCase.java) are broken today too, and your change fixes those as well — the inline list has no CamelCase rule at all. math.go staying false on both is the check that the swap doesn't over-match ordinary sources into "tests".

The one gap: your two new tests are both Python, so nothing here would catch a Go regression. If you'd like, I can send a commit adding math.go / math_test.go to __tests__/cli-affected-paths.test.ts against your branch — say the word and I'll open it, or ignore this if you'd rather keep the PR as-is. Not trying to take it over; it's your fix and it's been sitting for three weeks.

[Jseca666]

Thanks for the careful validation and for connecting this to #1507. Please go ahead and add math.go / math_test.go regression coverage to this branch — I agree it should be included before merge.

[Jseca666]

Hi @maxmilian, thanks again for offering to add the Go regression coverage. Just checking in — are you still planning to send the follow-up commit for math.go / math_test.go? No worries if timing is tight; I’m happy to add it on my side instead. Please let me know if you need anything from me to help move this forward.

[Jseca666]

Hi @colbymchenry, I noticed #1507 was closed via #1803, based on #1688.

My PR #1273 has been open since July 13 and already proposed replacing the duplicated test-file detection. On August 5, @maxmilian verified that it also fixes #1507 and linked it in both discussions.

Could you review this earlier contribution and clarify how it should be acknowledged? My PR also includes a separate --filter glob fix that appears to remain outstanding. I’d like to get that contribution merged—would you prefer that I update #1273 to focus on the remaining fix?

Thanks.