test(mcp): audit and expand Enterprise MCP live-integration coverage

[joshrotenberg]

Summary

Enterprise MCP currently has good mock coverage in crates/redisctl-mcp/tests/enterprise_tools.rs, but it still lacks the kind of live-cluster integration coverage we now have on the CLI side.

We should add a Docker-backed Enterprise MCP integration matrix against the local Enterprise demo cluster, biased toward safe read paths first and then selected write paths where cleanup is practical.

Why

The mock suite is useful for response-shape and serialization regressions, but it will not catch:

• wrong live endpoint wiring
• version-specific behavior differences
• response payload drift from a real Enterprise cluster
• commands that look fine under mocks but fail against the demo cluster

Proposed scope

Start with read-heavy MCP coverage across:

• cluster
• databases
• observability
• rbac
• proxy
• services

Then add carefully bounded write-path coverage where cleanup is straightforward.

Proposed approach

• Reuse the local Enterprise demo cluster already used by the CLI integration suites
• Add shared MCP-side Enterprise test support under crates/redisctl-mcp/tests/support/
• Add an ignored smoke matrix similar in spirit to the CLI Enterprise smoke work
• Prefer broad endpoint reach first, then deepen behavior assertions for the highest-risk commands

Acceptance criteria

• A Docker-backed Enterprise MCP integration suite exists
• Safe read paths cover the main Enterprise MCP submodules
• The suite is easy to run locally with the existing Enterprise demo stack
• Live failures found during the matrix are split into follow-up bug issues as needed

Context

Related to #628, #679, and #681.