Allow multiple authorized instances of the same connector (per-account, namespaced)

### Problem Statement

A connector can be authorized for one account per provider. People who legitimately operate across several accounts of the same provider can reach exactly one of them.

I run two companies and hold three Microsoft accounts across two tenants. The OneDrive desktop client handles all three simultaneously — the platform imposes no such limit. The Microsoft 365 connector allows one, so it is authorized for the first tenant only.

**Why this costs more than convenience.** For the authorized tenant, an agent filing a document writes through Graph and gets an item id back — server-side proof the file arrived. For the second tenant there is no API path, so the only route is the local sync folder, which fails silently when the sync client is not running. I hit exactly that: files appeared filed while nothing had been uploaded.

Losing the API for the second account does not cost convenience. It costs the only verifiable write path.

### Proposed Solution

Allow more than one authorized instance of the same connector, each with a user-chosen label, with tools namespaced per instance so the agent can address them explicitly.

This is not Microsoft-specific. The same gap applies to Google personal plus Workspace, multiple Slack workspaces, and multiple GitHub organizations. Anyone running more than one company hits it.

### Alternatives Considered

- **Cross-tenant folder sharing.** Exposes content without necessarily granting write access, and means giving one company visibility over another company's material. Not acceptable here.
- **Reorganizing shared content in place.** Graph's move operation does not cross drives, so this does not work even where sharing is in place.
- **Re-authorizing the connector per task.** Manual, and cannot be scripted or scheduled — which rules out any unattended or recurring workflow.
- **Falling back to the local sync folder for the second tenant.** This is what I do today, and it is the failure mode described above: it reports nothing, so a file that was never uploaded looks filed.

### Area

MCP Connector (adding/managing servers)

### Priority

High - Major improvement needed

### Additional Context

Related: #867 and #868, both on constraints in the same connector.

The overlap with #867 is worth noting: both issues end at the same place. Whenever the API path is unavailable — because of a size cap or because the account cannot be authorized — the fallback is a local folder that cannot report failure. The verifiable path is the one that disappears.

Start with the MCP Connector adding/managing servers area and read related issues #867 and #868. Done means multiple authorized instances of one connector can have user-chosen labels and explicitly namespaced tools; the payload names no files or tests, so implementation entry points and verification still need to be located.

Scritto dal modello di indicizzazione a partire dal testo della issue.