Configure per-user rate limits on MCPServer

[jerm-dro]

User Story

As a cluster admin,
I want to configure per-user rate limits on MCPServer,
so that no single user can monopolize a server's tools.

Context

See THV-0057: Rate Limiting for MCP Servers for full design details.

Acceptance Criteria

• MCPServerSpec includes a rateLimiting.perUser field with maxTokens and refillPeriod
• Per-tool per-user limits supported via rateLimiting.tools[].perUser
• CRD admission validation rejects perUser when auth is not enabled
• CRD admission validation rejects rateLimiting when neither global nor perUser is set
• Operator sets RateLimitingConfigValid status condition at reconciliation time
• Unit: User exceeds maxTokens — next request is rejected with JSON-RPC error -32029 and retryAfterSeconds
• Unit: User A is rate-limited; User B's bucket is independent
• Unit: Per-tool limit hit on expensive_tool doesn't affect other_tool
• Unit: Request must pass both server-level per-user and per-tool per-user limits
• Unit: Redis unavailable — request passes through (fail-open)
• E2E: Deploy MCPServer with perUser rate limit, send traffic, verify rejection after limit exceeded
• E2E: Streamable HTTP rejection returns HTTP 429 with Retry-After header
• E2E: kubectl apply with perUser and no auth config is rejected at admission

Out of Scope

• Global rate limits (STORY-002)
• VirtualMCPServer (STORY-003)
• Observability beyond fail-open counter (STORY-004)