refactor(bigquery-jdbc): optimize thread management and unify discovery logic in DatabaseMetaData methods (#13560)

b/520407325
b/520406763

This PR refactors and optimizes the database metadata retrieval methods
in the BigQuery JDBC driver. It resolves thread management
inefficiencies, eliminates duplicate API blocks, ensures consistent
project discovery support, and implements strict, fail-fast error
propagation across all asynchronous metadata methods.

## Key Changes

### 1. Catalog-Based Routing in `getSchemas`
* **Synchronous Single-Catalog Path**: If a specific catalog (project)
is requested, `getSchemas` now executes completely synchronously on the
calling thread, eliminating background executor and queue overhead.
* **Parallel Multi-Catalog Path**: If no catalog is specified, the query
executor runs parallel scans across all accessible projects (primary,
additional, and discovered) in the background.

### 2. Unified Dataset Discovery & Deduplication
* Introduced a single, shared helper method `fetchMatchingDatasets` to
serve as the sole entry point for listing and filtering datasets.
* Deduplicated dataset-listing logic across **7 metadata methods**
(`getSchemas`, `getTables`, `getColumns`, `getProcedures`,
`getProcedureColumns`, `getFunctions`, and `getFunctionColumns`),
ensuring consistent support for project discovery and SQL wildcard
matching across all of them.

### 3. Strict Fail-Fast Error & Interruption Handling
* **Preventing Silent Truncation**: Background metadata fetchers used to
silently swallow `InterruptedException`s and `ExecutionException`s,
returning partial or empty results to the user (which severely violates
the JDBC contract during query timeouts or connection drops).
* **Consistent Propagation**: All 7 background methods now strictly push
an `SQLException` to the client's queue whenever an interruption or
runtime error occurs, ensuring the user gets an explicit failure when
calling `rs.next()`.
* **Boilerplate Elimination**: Extracted the complex error-handling and
queue-finalization boilerplate into centralized helper methods
(`handleFetcherException` and `finalizeFetcher`). This removed over 50
lines of duplicate code, deleted 14+ redundant `isInterrupted()` checks,
and established a clean, unified standard for all asynchronous metadata
threads.

Author: keshavdandeva