bug(cli/mcp): find_callers collapses every row's target_file_path to the first row's, misattributing callers across same-named functions #1530

tools/code_finder.py:1216-1221

find_callers hoists one row's target path into the response envelope, then strips the field from every row:

target_file_path = next((r.pop("target_file_path", None) for r in results), None)

next(...) consumes only the first generator item, so the envelope gets row 0's value — but the surrounding loop still pops the field off all remaining rows, discarding their (different) values.

When context is not supplied, who_calls_function matches Function {name: $function_name} across all files, so rows legitimately carry different target_file_path values.

Reproduction

Two result rows targeting /repo/models/user.py and /repo/models/order.py produce:

{
  "target_file_path": "/repo/models/user.py",
  "callers": [ ...both callers, under the one path... ]
}

Impact

Ask for callers of any common method name — save, run, handle, get — that exists in several files, and every caller is reported as calling the definition in whichever file happened to sort first. An agent acting on that reads or edits the wrong file, with nothing in the response signalling ambiguity.

Suggested fix

Either keep target_file_path per row (drop the envelope hoist), or group results by target path and return a list of {target_file_path, callers} groups so genuine ambiguity is visible to the caller.

_{Reproduced by driving the result-shaping code with two rows carrying distinct target paths.}