Stop creating unused pmeta streams on query-server startup

[VibeSWE]

Upstream c7dfbefe removes automatic creation and synchronization of pmeta during query-server startup. The earlier OSS change 350ab9c8 removed its ingestion path, leaving unnecessary provisioning in the fork.

I propose adopting the startup cleanup in VibeEngineeringAI/parseable:

• Remove the query-server startup call and its creation/synchronization helper.
• Retain LogSource::Pmeta, the stream-name constant, and legacy migration/deserialization support.
• Preserve all existing stored metadata and data. This proposal does not delete or rewrite existing pmeta streams or change their existing access rules.

The patch applies cleanly to the fork's reviewed main commit bd20b043. That is a source-compatibility check; distributed runtime behavior still needs validation.

Before merging, verify:

1. Query-server startup against empty object storage does not create or sync pmeta.
2. Restarting an existing installation with pmeta metadata still loads successfully.
3. Query and ingestor startup works with both ready and temporarily unavailable ingestors, and ordinary ingestion/querying still works.
4. Legacy metadata migration remains supported.

Separately, the native dataset list and selectors could hide internal datasets by their authoritative stream type, with an intentional administrative view when needed. The existing Prism home API already supports includeInternal; the native UI can reuse that capability. This would cover existing pmeta and pstats without removing stored data. Hiding a dataset in the UI is not an authorization change, and can be reviewed independently.

Does any deployment still rely on automatic pmeta provisioning? Should internal datasets be hidden from normal exploration by default?