Expose an agent-readable index manifest

[tulayha]

Feedback from: This Twitter comment

Agents can inspect media readiness through get_workspace and operation schemas through get_capability, but the detailed identity of an active index is spread across internal snapshot and generation records.

Expose one concise, versioned view so clients can understand what they are searching without reading VidXP's database, desktop files, or internal manifests.

Expected behavior:

• Build the response from the existing media, snapshot, generation, and capability records; do not create a second source of truth.
• Include the repository and active snapshot identity, each media item's source hash and generation, indexed capabilities, record counts, and the model or configuration used to build them.
• Include how the transcript was obtained when that information is available, such as generated speech recognition or an imported transcript.
• Report whether an entry is current, superseded, or needs reindexing, with a reason. Do not rely on an arbitrary expiry date.
• Keep large record or segment ID lists behind paginated detail operations rather than returning the whole index at once.
• Derive available operations from the current authenticated session and MCP tool discovery instead of storing a static allowed-tools list.
• Expose the same application response through MCP and HTTP, with an optional concise CLI view.
• Do not expose server filesystem paths, credentials, or internal storage locations.

The initial implementation is useful even if it covers only the active snapshot and media generations, provided the schema can be extended without breaking clients.