fix(provider): report null vendor as missing in the provider registry

[yyqdbngt]

Summary

InstanceProviderRegistry.forVendor and catalogFor looked the vendor up
directly in an EnumMap. EnumMap.get(null) throws NPE, so a null vendor
crashed with a raw NullPointerException instead of the same 501
"No ... provider registered for vendor" that an unregistered vendor already
produces. Both lookups now map a null vendor onto the existing missing-entry
code path.

Why

The registry is the single lookup SPI for vendor-scoped providers; current
call sites normalize a null InstanceVO.vendor to APACHE first, but the SPI
contract ("unknown vendor -> 501") should hold for null too, so a future
caller passing a raw persisted vendor cannot 500 the request with an NPE.

Testing

cd server && mvn -Dtest="InstanceProviderRegistryTest" test

Tests run: 12, Failures: 0, Errors: 0, Skipped: 0
New regression tests forVendorShouldReportNullVendorAsMissingTest and
catalogForShouldReportNullVendorAsMissingTest assert the 501 BusinessException
(message "No ... provider registered for vendor null") instead of an NPE.

[RockteMQ-AI]

LGTM. Correct guard against EnumMap.get(null) NPE — surfacing a 501 for null vendors is the right behavior. Tests verify both forVendor and catalogFor.

---

Automated review by github-manager-bot