Force single worker when MCP is mounted

[wayyoungboy]

Refs #1056

Note: this PR implements the short-term single-worker guard from #1056. The longer-term stateless MCP session handling needed for true multi-worker support remains tracked there.

Summary:

• Detect when the server CLI can mount the streamable HTTP MCP app.
• Reuse the existing single-worker guard so MCP sessions are not split across multiple worker processes.
• Keep multi-worker startup unchanged when no process-local runtime mode is active, and avoid running the guard checks in reload mode.

Validation:

• Using the project virtual environment, ran python -m pytest tests/unit/server/test_server_cli.py tests/unit/server/test_main_mcp_optional.py -q (31 passed, 1 warning from an existing dependency deprecation).
• Using the project virtual environment, ran python -m py_compile src/server/cli/server.py tests/unit/server/test_server_cli.py.
• Ran git diff --check.
• E2E: with MCP dependencies installed and non-embedded storage configuration, started powermem-server --host <loopback host> --port <local test port> --workers 2 --no-open-browser; startup printed the MCP single-worker guard and forced workers=1 before server initialization. The process was stopped after the guard was observed.

[lightzt99]

Prefer longer-term stateless MCP mentioned in #1056 after discuss.

[wayyoungboy]

Self-review status for the current head: I reviewed the single-worker guard for MCP-mounted server runs. The change matches the current mount condition: when fastmcp is available, the server can mount the MCP ASGI app, so forcing workers=1 avoids running that mounted MCP app in multiple worker processes. Reload mode remains untouched, and the current checks are green.

I did not find a blocking issue in this static review. I am not marking this LGTM because this PR is authored from the wayyoungboy branch and still needs independent maintainer review/acceptance.