Fix embedded DB lock contention by closing nocache DoltDBs and plumbing DBLoadParams through CREATE DATABASE / clone / stats

[coffeegoddd]

Summary

This change fixes an embedded / driver reopen failure where a two-phase flow (CREATE DATABASE then later Ping) could still
hit the database is locked by another dolt process even with disable_singleton_cache + fail-fast lock behavior enabled.
Key updates:
• Close underlying DoltDBs in nocache mode: sqle.DoltDatabaseProvider.Close() now closes all underlying *doltdb.DoltDB
instances when DisableSingletonCacheParam is in effect, ensuring .dolt/noms/LOCK is released on engine shutdown.
• Plumb DBLoadParams into provider-managed DB creation paths:
• CREATE DATABASE / UNDROP DATABASE now use env.LoadWithoutDB so DBLoadParams can be applied before any DB is opened.
• dolt_clone applies provider DBLoadParams to the clone env before registration.
• registerNewDatabase defensively applies provider DBLoadParams to the env.
• Propagate DBLoadParams into stats backing store: statspro now uses LoadWithoutDB for new stats storage and copies
DBLoadParams from the host env so embedded-mode settings apply consistently.
• Regression coverage: adds a test that creates a DB via the SQL engine, closes the engine, and asserts the LOCK is
released.

Why

Embedded callers rely on deterministic reopen semantics under contention. Previously, the SQL provider close path didn’t
close the underlying *doltdb.DoltDB for newly created DBs in nocache mode, leaving the journal manifest LOCK held and
causing retries / reopens to fail.

[coffeegoddd]

@coffeegoddd DOLT

comparing_percentages
100.000000 to 100.000000

version	result	total
80225a3	ok	5937471

version	total_tests
80225a3	5937471

correctness_percentage
100.0