doctor: requests-proxy check — Part 2: real egress probe (Part 1 framing shipped in #356)

[LukasWodka]

Part of epic tracebloc/backend#1142 (Themes B/E, cli side).

MED. checkRequestsProxy (doctor.go:368-387) has three correctness issues:

1. False green: treats the Deployment's ReadyReplicas>0 as "Service Bus egress works" (doctor.go:379) — readiness ≠ the proxy can actually relay to Service Bus.
2. Wrong direction: detail/remedy say requests-proxy-down means "experiments stay Pending" (doctor.go:367,376,384). requests-proxy is on the OUTBOUND result/weights path — a down proxy stalls result egress mid-run, it does NOT block scheduling.
3. No real egress probe: the ✔ is inferred, never probed (the WS3 in-cluster SB-reachability probe noted at doctor.go:329-331 is still a TODO).

Fix: reword to "result/weights egress" (not scheduling); downgrade the inferred ✔ to reflect "deployment ready (egress not directly probed)"; and add the in-cluster SB-reachability probe as the real signal. Pairs with the client-chart readiness-probe ticket.

[LukasWodka]

Part 1 (result-egress framing + honest check wording) shipped in #356 (merged). Remaining Part 2: replace the readiness-based false-green with a real egress probe that verifies the proxy can actually relay to Service Bus (readiness ≠ egress works).