fix: clean up ServerlessRuntime record and cache on plugin uninstall

[GareArc]

Summary

• Delete ServerlessRuntime DB record when the last tenant uninstalls a serverless plugin (Refs==0), inside the existing transaction — guarded by InstallType == SERVERLESS
• Expose ClearServerlessRuntimeCache as a public method on PluginManager (renamed from private)
• Clear Redis serverless runtime cache on full plugin deletion — best-effort (logged, not returned), guarded by RuntimeType == SERVERLESS

Fixes ENG-474: after uninstall + reinstall, the stale ServerlessRuntime record and cached hostname caused credential requests to route to the deleted serverless endpoint (no such host). Required a Pod restart to recover.

Test Plan

• Install serverless plugin (single tenant) → uninstall → confirm serverless_runtimes row deleted and redis-cli GET "serverless:runtime:<id>" returns (nil)
• Reinstall same plugin → confirm fresh record created → add model credentials → no no such host error, no Pod restart needed
• Multi-tenant: install on Tenant A + B → Tenant A uninstalls (Refers → 1) → confirm record and cache NOT cleared (Tenant B still active)

[gemini-code-assist]

Code Review

This pull request exports the ClearServerlessRuntimeCache method and updates the plugin uninstallation flow to clear the serverless runtime cache and delete the corresponding ServerlessRuntime database records. The reviewer identifies potential cache and database record leaks in the UpgradePlugin flow, which mirrors the uninstall flow, and suggests logging an error if the plugin manager is uninitialized when attempting to clear the cache.