feat(admin): classify RustFS diagnostic capabilities

[cxymds]

Related issue

Closes rustfs/backlog#1403
Parent: rustfs/backlog#1383
Roadmap: rustfs/backlog#1361

Background and user impact

RustFS 1.0.0-beta.10 registers several diagnostic routes that are placeholders rather than real measurements. Route presence and HTTP 200 responses are therefore not sufficient evidence of support. Without probe-specific truth, rc could advertise or execute object, network, site, or site-replication performance tests that return misleading success.

The existing /v3/inspect-data endpoint also returns reconstructed object bytes and is not an inspect archive.

Root cause

Runtime discovery exposed broad server capabilities but did not normalize individual diagnostic probe states or provide a reusable fail-closed guard. Unknown future server versions could not be distinguished safely from the pinned beta.10 contract.

Solution

• Add 10 stable diagnostic capability names.
• Classify the pinned beta.10 contract:
  • available: health, cluster snapshot, extensions catalog, drive observations, and client-devnull
  • unsupported: inspect archive
  • stubbed: object, network, site, and site-replication netperf
• Mirror cluster and extension availability from their actual read-only discovery requests.
• Treat unknown future versions as unknown and missing v4 discovery as version-gated.
• Preserve permission-denied, disabled, unsupported, stubbed, version-gated, and unknown states.
• Add a reusable CapabilityReport guard that permits only explicitly available diagnostics.
• Verify discovery never executes an active diagnostic or speed-test request.

Validation

Passed locally at exact head f6e9316:

• cargo fmt --all --check
• cargo clippy --workspace --all-targets -- -D warnings
• cargo test --workspace

Focused rc-core and rc-s3 capability tests also pass.

Review notes

This PR is intentionally stacked on #268 because it extends runtime capability discovery and output schema v3 foundations. It adds no diagnostic command and performs no active probe. Unknown versions remain fail-closed until a server-owned per-probe contract is available through rustfs/backlog#1402.

BREAKING process

BREAKING marker for the stacked exact-head workflow: temporary validation against main includes protected command-reference/schema changes from prerequisite PRs #266 and #268. This #1403 diff against its true #268 base changes only capability core/S3 code and does not independently modify a protected file. The prerequisite output-v3 stack carries its documented additive contract and migration boundary.

[cxymds]

Validation complete at exact head f6e9316.

• CI: https://github.com/rustfs/cli/actions/runs/29845541582 (passed on attempt 2)
• Integration Tests: https://github.com/rustfs/cli/actions/runs/29845541530 (passed)
• Local fmt, all-target Clippy, and workspace tests passed.

The first advisory attempt failed before checkout while Docker Hub timed out resolving the pinned cargo-deny action base image. Re-running only the failed job succeeded without a code or dependency change.

The PR remains correctly stacked on #268 and performs no active diagnostic request.