Unblock main CI: fix vMCP nil-registry panic and CRD docs drift

[jhrozek]

Main is currently red on two independent checks, both fallout from the "Handle per-backend MCP revision in vMCP" merge. Since every PR branches off main, a fix for just one still inherits the other's red and never gets a green run — so this PR clears both.

1. Tests / Test Go Code — nil-registry panic

TestRegression_ToolSchemaFidelity_PreservesCompositors panics with a nil-pointer dereference, and because a panic tears down the whole pkg/vmcp/client test binary, every test in the package reports as failed.

Why: the test builds httpBackendClient as a raw struct literal that never sets registry. The per-backend revision handling made ListCapabilities run probeRevision -> buildModernHTTPClient -> resolveAuthStrategy, which calls h.registry.GetStrategy(...) on the nil registry (client.go:454). The real constructor NewHTTPBackendClient rejects a nil registry, so this is a test-setup gap, not a missing production guard.

What: build the probe client via the existing newProbeClient(t) helper (wires a real registry with UnauthenticatedStrategy). Test-only.

2. Operator CI / Generate CRD Docs — stale CRD reference

The same merge added an MCPRevision field to the backend status but did not regenerate docs/operator/crd-api.md, so crd-ref-docs + git diff --exit-code fails. Regenerated to add the missing mcpRevision entry. Generated-docs only.

Both are carved from the fixes bundled in #5993 (go-sdk v1.7 migration) so main and dependent PRs can go green now instead of waiting on that larger PR. cc @JAORMX — #5993 can rebase cleanly over this.

Type of change

• Bug fix (non-breaking change which fixes an issue)

Test plan

• Unit tests (task test): go test ./pkg/vmcp/client/ -run TestRegression_ToolSchemaFidelity_PreservesCompositors -count=2 — panics before, passes after.
• task crdref-gen produces no further diff after this change.

Does this introduce a user-facing change?

No.

Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 72.11%. Comparing base (885e8b1) to head (9d794c2).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #5999      +/-   ##
==========================================
+ Coverage   72.06%   72.11%   +0.04%     
==========================================
  Files         718      718              
  Lines       74462    74462              
==========================================
+ Hits        53662    53698      +36     
+ Misses      16965    16912      -53     
- Partials     3835     3852      +17     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[JAORMX]

LGTM — this is the right fix for the main breakage, and the diagnosis matches exactly what I traced when #5980 + #5990 landed together: the regression test built httpBackendClient as a bare struct literal with a nil registry, and the post-#5980 ListCapabilities → probeRevision → buildModernHTTPClient → resolveAuthStrategy path derefs h.registry.GetStrategy(...) (client.go:454) → nil panic that tears down the whole pkg/vmcp/client test binary (so every test in the package, and the Test Go Code gate, goes red on main and every open PR).

Routing through newProbeClient(t) (real registry + UnauthenticatedStrategy) is the correct, minimal, test-only fix — and correctly framed as a test-setup gap, not a missing production guard, since NewHTTPBackendClient already rejects a nil registry. Overriding h.clientFactory afterward preserves the test's streamable-HTTP backend. Good call carving this out of #5993 so main can go green now.

Approving. Worth merging promptly since it unblocks CI across all open PRs.

[JAORMX]

Re-approving after the scope expansion (head 9d794c2). Good call folding the second main-red check in — both failures are #5980 fallout and, since every branch forks off main, fixing only one would still inherit the other's red, so clearing both together is the right move.

• Nil-registry panic (Test Go Code): unchanged from my earlier approval — routing through newProbeClient(t) is correct.
• CRD docs drift (Generate CRD Docs): the one-line mcpRevision entry in docs/operator/crd-api.md matches the MCPRevision field #5980 added to the discovered-backend status, and you confirmed task crdref-gen yields no further diff. Generated-docs only.

Still test/docs-only, no production change. LGTM — worth merging promptly to unblock CI across all open PRs.