fix(provider): honour driver_name when a connection string omits the DBAPI

[C1-BA-B1-F3]

Summary

A provider configured with a bare connection string:

data: postgresql://postgres:secret@localhost/test

resolves to SQLAlchemy's default DBAPI for the backend — psycopg (v3) for PostgreSQL. Deployments that only install psycopg2 (per requirements-provider.txt) then fail at engine creation:

sqlalchemy/dialects/postgresql/psycopg.py in import_dbapi
ModuleNotFoundError: No module named 'psycopg'

even though PostgreSQLProvider.__init__ pins driver_name='postgresql+psycopg2'. The pin is silently ignored whenever data is a string, because store_db_parameters routes it to self.db_conn and get_engine uses the URL verbatim.

The CI failure in tests/provider/test_postgresql_provider (and test_api_connection_rfc3986 in the manager suite, which also builds its connection string with a bare postgresql:// URL) is exactly this.

Fix

In get_engine, when the caller's URL has no +driver suffix and the provider's driver_name does, rewrite the URL's drivername to the pinned one. Explicitly-pinned URLs (postgresql+psycopg://...) are untouched.

Test plan

• Verified with mocked create_engine:
  • postgresql://… + pin postgresql+psycopg2 → engine URL becomes postgresql+psycopg2
  • postgresql+psycopg://… → untouched
  • no conn_str → built from driver_name as before
• test_api_connection_rfc3986 now reaches the real connection attempt (psycopg2.OperationalError: connection refused locally) instead of failing at import psycopg, confirming the v2 driver is selected.

Fixes #2424

[C1-BA-B1-F3]

The Build failure is in test_describe_collections (assert 9 == 10) and test_get_collection_edr_query (HTTP 500), the same two pre-existing upstream regressions that also fail on the current master branch and on my other PRs (#2427, #2428). They exercise /collections and EDR queries; this PR only touches pygeoapi/provider/sql.py (the get_engine helper). Happy to rebase once upstream fixes the CI environment.