feat(search): add configurable file-level deduplication

[Don-Yin]

problem

when a search query matches several chunks from the same file, the result list fills up with duplicate file entries. this crowds out results from other relevant files and reduces the effective diversity of the top-N list. in practice, a user who asks for 10 results may receive 4 or 5 entries that all point to the same file, which leaves fewer slots for other files that would otherwise rank highly.

solution

this PR adds a search.dedup.enabled option (default: true) that retains only the highest-scoring chunk per file path. deduplication runs after boost scoring, so the best chunk is selected with structural penalties and bonuses already applied.

when dedup is enabled, the internal fetch limit increases from limit * 2 to limit * 4 so that enough unique files survive the filter. users can disable it in config.yaml if they prefer the original behaviour:

search:
  dedup:
    enabled: false

changes

file	what
config/config.go	add DedupConfig struct, wire it into SearchConfig, set the default to true
search/search.go	read the dedup config, scale fetchLimit conditionally, call DeduplicateByFile when enabled
search/dedup.go	implement DeduplicateByFile (keeps the first occurrence per file path from a pre-sorted slice)
search/dedup_test.go	unit tests: standard case, empty input, all-unique input

test plan

• go test ./search/ passes (all 20 tests, including 3 new dedup tests)
• go test ./config/ passes
• go build ./... succeeds with no warnings
• manual verification: run grepai search <query> --json and confirm that each file path appears at most once in the output
• manual verification: set search.dedup.enabled: false in config.yaml and confirm that duplicate file entries reappear

note

this branch will remain open for additional work (e.g. wiring dedup into the workspace search path). further commits may follow.