fix(grep): dedupe overlapping context windows; honor max_matches=0

[t957095]

Both halves of #165:

1. Overlapping context windows duplicated lines. Each match appended its full window with no tracking — a 5-line file with 3 matches and context=2 emitted 11 lines. A per-file printed set now emits each line at most once.
2. max_matches=0 returned one match labelled (capped at 0). The cap was checked after appending; the check now runs before a match is emitted, so 0 yields no matches and max_matches=1 still returns the first hit plus the cap note.

Tests: three regression tests added (overlap dedupe emits each line once and in order; max_matches=0; max_matches=1 keeps the first hit). All pass.

Note: test_grep_context_lines_returns_neighbors and test_grep_context_zero_is_default_and_no_context_markers fail on Windows on clean main too — _rel() emits OS-native separators (src\a.py vs the asserted src/a.py). Pre-existing, unrelated to this change; happy to send a small follow-up normalizing separators if you want.

Closes #165

[Copilot]

Copilot review overview

🟡 Changes recommended

Overlapping matches can retain an incorrect context marker, and the regression test does not assert marker correctness.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1

Open (1)

• Preserve match marker when a context line later matches · New

What changed in this PR

Updates grep to deduplicate overlapping context output and correctly honor max_matches=0.

Changes:

• Deduplicates overlapping context lines.
• Checks match caps before emitting results.
• Adds regression tests for overlap and match limits.

File	Description
tests/​test_tools.py	Adds regression coverage for deduplication and match caps.
src/​opendot/​tools/​local.py	Implements context deduplication and cap handling.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[t957095]

Addressed the Copilot finding in the latest commit: when an overlapping context window reaches a line already emitted as context and that line is the current match (j in printed and j == i), the entry is upgraded from the context dash to the match colon instead of keeping the stale marker — the fixture now reports f.txt:3:aaa, not f.txt:3-aaa.

The overlap regression test asserts every marker explicitly: : / - / : / - / :. All grep tests pass locally; the two remaining context-test failures on my Windows box are the pre-existing _rel() path-separator issue noted in the commit message, and fail identically on unmodified main.

[vedaant00]

Nice work @t957095, logic is correct on all three counts, and I like that the overlap test asserts every marker rather than just counting lines. Verified locally: output now matches real grep -C2 exactly, max_matches=0 returns no matches, and cross-file line collisions are handled. Both new tests fail on pre-fix code and pass with it.

Only the lint job was red (three lines over the width limit). I pushed a ruff format commit to your branch, no behavior change. Good to merge once CI is green.

And yes please on the Windows _rel() separator follow-up, that's a real gap our Linux-only CI can't see.