find_source_in_db's spatial fallback search crashes on a completely empty Sources table

**Where:** `astrodb_utils.sources.find_source_in_db` (via `ingest_source` with `search_db=True`), called from the `astrodb-ingest-sources` skill's generated script.

**What happened:** Ingesting the very first source into a brand-new, completely empty `Sources` table with `search_db=True` failed on every single row with `"None of [Index(['ra_deg', 'dec_deg'], dtype='str')] are in the [columns]"`. This comes from `find_source_in_db`'s coordinate-based `db.query_region(...)` fallback (reached because there's no name match yet), which appears to return a DataFrame with no columns at all when queried against a table with zero rows, so the subsequent `db_name_matches[ra_col_name]` access fails.

**Workaround:** Set `search_db=False` for this first bulk ingest into a fresh table — there is nothing to dedupe against yet, so skipping the search entirely is correct and cheap. This would need to be revisited (`search_db=True`) for any future incremental ingest into an already-populated `Sources` table.

**Suggested change:** `astrodb-ingest-sources`'s skill instructions could note this edge case explicitly: "if this is the first ingest into an empty database, prefer `search_db=False`, since `find_source_in_db`'s spatial-search fallback errors on a zero-row table rather than returning zero matches." Longer-term, the fix belongs in `astrodb_utils.sources.find_source_in_db` (guard the empty-table case before indexing into `db_name_matches`).

---
_Reported from a gotchas.md log filed by a skill user (2026-08-28)._

Start at `astrodb_utils.sources.find_source_in_db`, reached through `ingest_source` and the `astrodb-ingest-sources` generated script, and reproduce the first ingest against an empty `Sources` table with `search_db=True`. Confirm that the spatial fallback handles zero rows without a column-indexing crash, then check the skill instructions for the documented first-ingest workaround. Done means the empty-table case is handled or clearly documented without disrupting incremental searches.

Written by the indexing model from the issue text.