Delete TestMatchesGlobs and TestMatchesAction private function tests

[gregology]

Fixes #76

Our docs/testing.md literally calls out matchesGlobs by name as the canonical example of what not to test: "if matchesGlobs breaks, a Resolve test will catch it. Testing both just means two tests to update when the signature changes." And yet there it was, sitting in the test suite alongside TestMatchesAction, doing exactly the thing the docs tell contributors not to do.

This matters because new contributors read the tests before they read the docs. Having private function tests in the codebase signals "this is how we do things here" regardless of what the guidelines say. It undermines the whole testing philosophy we spent time documenting.

Every scenario these two tests covered is already exercised through the public Resolve() API:

• Glob matching (double-star, extensions, directory-scoped, single-star) -> TestResolve_NonPythonFile, TestResolve_ExcludePattern, TestResolve_EditBefore
• Exclude overrides -> TestResolve_ExcludePattern + the ExcludeOverridesMatch property test
• Directory boundaries -> TestResolve_ContextAboveRootIsIgnored
• Action filtering -> TestResolve_ReadDoesNotMatchEditOnly, TestResolve_CreateAction, and the EditEntriesFilteredForActionRead property test

No coverage gaps. Just removed the duplication.

Why not keep them as "extra safety"? Because that's exactly the trap the docs warn about. Two tests covering the same behavior means two things to update when internals change, which makes refactoring harder and trains people to see private function tests as normal. The 19 remaining TestResolve_* tests (including 5 property-based ones with randomized inputs) give us better coverage than the deleted unit tests ever did.

Dev Ghost metrics

Total duration: 4m 32s
Turns: 60
Tool calls: 48
Tokens: 368,976 input / 5,084 output

Stage	Model	Duration	Turns	Tool calls	Tokens (in/out)	Cache read	Cache creation
triage	claude-opus-4-6	1m 18s	27	25	40,107 / 1,097	34,776	5,327
implementation	claude-opus-4-6	2m 3s	14	9	135,650 / 2,482	114,256	21,382
evaluation	claude-opus-4-6	0m 20s	4	2	39,844 / 428	35,559	4,281
docs_review	claude-opus-4-6	0m 30s	14	11	92,512 / 527	69,616	22,890
craft_pr	claude-opus-4-6	0m 19s	1	1	60,863 / 550	35,641	25,219

Resolves #76