Add ucode configure --mcp one-shot flag

[sunishsheth2009]

Summary

Adds a --mcp flag to ucode configure so an agent and its Databricks MCP server(s) can be set up in a single command:

ucode configure --agents claude --mcp system.ai.slack

Previously this took two steps (ucode configure --agents claude then ucode configure mcp --services system.ai.slack). The new flag folds them together for a cleaner getting-started story (e.g. the AI Gateway MCP-services "Get started" panel can now recommend one command per client).

Behavior

• --mcp takes a comma-separated list of fully-qualified <catalog>.<schema>.<name> service names.
• Runs after the workspace + agents are configured (right beside the existing --tracing hook) and dispatches to the already-tested configure_mcp_command(services=...).
• Without --agents (e.g. MCP-only clients such as Cursor): configures just the workspace — no interactive agent picker — then registers the servers.
• Bare short names (slack) are rejected with a pointer to ucode configure mcp for the interactive picker.

Changes

• src/ucode/cli.py: --mcp option on the configure callback + handler; an MCP-only branch that does a workspace-only configure_shared_state when no --agent(s) is given.
• tests/test_cli.py: TestConfigureMcpFlag — with-agents dispatch, MCP-only workspace-config path (no picker), bare-short-name rejection.
• README.md: documents the flag.

Verification

uv run pytest → all CLI tests pass (tests/test_cli.py: 83 passed); uv run ruff check clean. Verified against a live workspace that the MCP-only path resolves clients correctly ("Configuring for: Cursor") and queries the real Unity Catalog MCP-services API before registering.

Note: the Cursor MCP-only path pairs with the ucode cursor client support in the separate cursor PR; the --mcp flag itself is independent and works today for the model agents.

This pull request and its description were written by Isaac.

[sunishsheth2009]

Note: this flag is transport-independent — it dispatches to configure_mcp_command, which under #201 registers each client's Databricks MCP server as a ucode mcp-proxy stdio entry. Verified it merges cleanly on top of #201 (no conflict) and its tests pass on that base, so it can land in either order relative to #201.

[sunishsheth2009]

Rebased onto latest master — conflict resolved (GitHub now reports MERGEABLE).

The only conflict was in cli.py: master added the --enable-fable and --enable-databricks-ai-tools configure options in the same option block where this PR adds --mcp. Resolved by keeping all three. No logic changes to the --mcp behavior.

This PR is independent of the proxy PR (#201) — the --mcp flag dispatches to the existing configure_mcp_command(services=...) that's already on master, and touches only cli.py / README.md / test_cli.py. It can merge on its own.

Verified locally after rebase: ruff format --check + ruff check + ty check clean; 920 passed (the 2 test_e2e_user_agent failures are the pre-existing live-gateway network tests, unrelated to this change and failing identically on master).