fix(query): pipeline resolver — handle column-keyed row objects

[alexskatell]

Why

Follow-up to #12. The v1.6.2 pipeline-name resolver shipped with a row-shape mismatch that made it silently fail against the real hosted warehouse:

$ topline --agent query audit --pipeline 'flex triage' --since this-week-et
no pipeline matched "flex triage". Available pipelines: (none found)

SELECT id, name FROM pipelines returns rows fine; the resolver simply didn't recognize the row shape.

Root cause

The hosted warehouse query API returns rows as column-keyed maps:

{"columns": ["id","name"], "rows": [{"id":"bna6...","name":"Sales - Flex - Triage"}]}

But selectIDNamePairs assumed positional arrays ([]any aligned with columns). Every row failed the r.([]any) type assertion and got silently skipped, leaving the resolver to report "(none found)".

The unit-test fixture in pipelineLookupServer also emitted positional arrays, so the mismatch slipped through CI.

Fix

• selectIDNamePairs now switches on row type:
  • map[string]any → read row["id"] / row["name"] by column name.
  • []any → read by positional index (preserved for parity with any deployment that returns array rows).
• Test fixture rewritten to emit the real column-keyed object shape — resolver tests now exercise the actual wire format.

Verified live

$ topline --agent query audit --pipeline 'flex triage' --since this-week-et --status open
{ "activity": { ... rewritten_sql: "...WHERE \"pipeline_id\" = 'bna6e9DoPgRchNsjeYS3'..." ... } }

Resolves cleanly to bna6e9DoPgRchNsjeYS3 (Sales - Flex - Triage) and runs the audit.

All existing resolver tests still pass.