fix: Merge extra on update instead of replacing

[mashawes]

Fixes update operations so the extra JSON field is merged with the existing value instead of being fully replaced, matching Python Keystone behavior

• Adds a shared merge_extra helper in crates/core/src/db.rs: the currently stored extra object is used as the base, keys in the update overwrite or add entries, and a key whose new value is null unsets that key. This preserves properties the caller did not resend while still allowing individual properties to be cleared.
• Wires the helper into all four driver update operations that handle extra: service, region, and endpoint in the catalog-sql driver, and user in the identity-sql driver. Each now reads the existing row's extra and merges rather than overwrites.
• Adds unit tests for merge_extra covering preserve, overwrite, add, null-unset, no-op null, and empty/missing existing blobs, plus an integration test under tests/integration/src/catalog/service/update.rs that creates a service with extra and verifies an update preserves an untouched key, adds a new one, and unsets a key via null.

Closes #777

Note: this contribution was developed with AI assistance.

[gtema]

The change on it's own is clean, there is however one critical issue: provider sends one value for update (which is also logged in traces), but drivers persist a different value. It also opens doors for situations that for 1 provider one drivers merges the data, while the second one misses it and overwrites the data. The same is valid across providers.

It must be responsibility of the provider to prepare the data, while the backend driver is only responsible for persistence.