extract_table_names should handle qualified table names (schema.table)

[ndenev]

Summary

extract_table_names mishandles qualified table references (schema.table), capturing only the first identifier after FROM/JOIN.

Affected code

• src/datafusion_integration/preprocess.rs (extract_table_names)

Root cause

The implementation takes the next Token::Word after FROM/JOIN and pushes it directly.
For qualified names like public.pods, token stream is effectively:

• Word(public)
• Period
• Word(pods)

Current code records public as the table name.

Why this matters

extract_table_names feeds table-aware JSON column detection (preprocess_sql_with_registry).
If table extraction is wrong:

• registry lookups miss actual table metadata,
• table-specific/custom JSON columns may not be recognized,
• dot-notation preprocessing can diverge from intended behavior.

Impact

Medium correctness/robustness issue, especially for qualified references and custom JSON columns.

Proposed fix

Preferred:

1. Use SQL AST extraction (from parsed Statement) to collect TableFactor object names and select the relation/table component correctly.

Minimal fallback:
2. Improve token-based extraction to consume dotted identifiers and use the last segment (catalog.schema.table -> table).

Tests to add

1. SELECT * FROM public.pods -> pods
2. SELECT * FROM postgres.public.pods -> pods
3. SELECT * FROM public.pods p JOIN public.services s ON ... -> pods, services
4. Quoted identifier variants where supported.