fix: over-fetch in find_usages so definition filtering doesn't shrink results

[GoodbyePlanet]

Fixes #74

Problem

find_usages asked the store for exactly limit fused hits, then filtered out any hit whose symbol_name equals the searched symbol — i.e. the symbol's own definition, which is not a usage.

Because the definition is almost always a top hit for the query "code that uses or references X" (it is the most semantically similar thing to its own name), the caller ended up with limit - 1 results even when more real usages existed just past the fetch window. At limit=1 it reported "No usages found" for symbols with plenty of usages.

Fix

Fetch limit + USAGE_OVERFETCH hits before filtering, then slice to limit.

The buffer is 5 rather than 1 because a single name can legitimately produce several definition hits — overloaded methods, or the same class name present in more than one indexed service.

Tests

tests/tools/test_search.py contained test_find_usages_can_undercount_when_all_fetched_hits_are_self_matches, which asserted the undercount as a documented known limitation. It is replaced by two tests:

• the store is queried with limit + USAGE_OVERFETCH
• a caller asking for 3 usages receives 3 even when the definition ranks first

Full suite: 293 passed.

🤖 Generated with Claude Code