Stop FastMCP from configuring application logging

[Genmin]

Fixes #1656.

Summary

• remove process-wide logging configuration from MCPServer.__init__
• configure logging in the mcp run CLI entrypoint instead, where the process is owned by the SDK CLI
• preserve server log-level settings for CLI-run servers and fall back to INFO for non-FastMCP server objects

Validation

• uv run ruff format --check src/mcp/cli/cli.py src/mcp/server/mcpserver/server.py tests/cli/test_utils.py tests/server/mcpserver/test_server.py
• uv run ruff check src/mcp/cli/cli.py src/mcp/server/mcpserver/server.py tests/cli/test_utils.py tests/server/mcpserver/test_server.py
• uv run pyright src/mcp/cli/cli.py tests/cli/test_utils.py tests/server/mcpserver/test_server.py
• uv run pytest tests/cli/test_utils.py tests/server/mcpserver/test_server.py -q
• git diff --check

[Genmin]

Follow-up pushed for the remaining CI coverage gate. The failed matrix jobs were passing tests but reporting two uncovered lines (src/mcp/cli/cli.py:126 and tests/server/mcpserver/test_server.py:54), so I added coverage for the invalid log-level fallback and replaced the local no-op callback with a MagicMock assertion that does not create an uncovered function body.

Local validation on the updated branch:

• uv run pytest tests/cli/test_utils.py tests/server/mcpserver/test_server.py tests/server/mcpserver/utilities/test_logging.py -q
• uv run ruff format --check tests/cli/test_utils.py tests/server/mcpserver/test_server.py tests/server/mcpserver/utilities/test_logging.py
• uv run ruff check src/mcp/cli/cli.py src/mcp/server/mcpserver/utilities/logging.py tests/cli/test_utils.py tests/server/mcpserver/test_server.py tests/server/mcpserver/utilities/test_logging.py
• uv run pyright src/mcp/cli/cli.py src/mcp/server/mcpserver/utilities/logging.py tests/cli/test_utils.py tests/server/mcpserver/test_server.py tests/server/mcpserver/utilities/test_logging.py
• git diff --check

[Genmin]

One more narrow follow-up pushed: the next Ubuntu run reached 100% coverage, then strict-no-cover correctly reported that the run() pragma was now stale because the CLI path is covered by the new test. I narrowed the pragma to the exception path only.

Additional local validation:

• uv run pytest tests/cli/test_utils.py tests/server/mcpserver/test_server.py tests/server/mcpserver/utilities/test_logging.py -q
• uv run ruff format --check src/mcp/cli/cli.py tests/cli/test_utils.py tests/server/mcpserver/test_server.py tests/server/mcpserver/utilities/test_logging.py
• uv run ruff check src/mcp/cli/cli.py src/mcp/server/mcpserver/utilities/logging.py tests/cli/test_utils.py tests/server/mcpserver/test_server.py tests/server/mcpserver/utilities/test_logging.py
• uv run pyright src/mcp/cli/cli.py src/mcp/server/mcpserver/utilities/logging.py tests/cli/test_utils.py tests/server/mcpserver/test_server.py tests/server/mcpserver/utilities/test_logging.py
• git diff --check

[smurfix]

Thank you. Is there anything that blocks this PR? I'd really like to remove the required monkeypatching of logging.basicConfig from my code …