test(compat): lossmaxttl baseline A/B calibration + gain-hunt scaffold

[andrescera]

Affected repo & language: srtla — C / compat test harness

What

Adds two compat-test additions backing the receiver-capability reconciliation:

1. A lossmaxttl 30-vs-40 baseline calibration A/B scenario (Task 1).
2. A scaffold for the gain-hunt adverse-config harness plus its protocol doc
   (Task 2).

Why

The reconciliation needs an empirical baseline winner. The A/B calibration
proves 30 and 40 tie at zero drop/errors, so the pre-registered tie-break
resolves to lossmaxttl=40 (BellaBox parity). The gain-hunt scaffold makes
future FEC/parameter gain experiments reproducible. See
docs/RECEIVER-RECONCILIATION.md and docs/GAIN-HUNT-PROTOCOL.md.

How to verify

• cmake -B build && cmake --build build then ctest.
• Compat matrix: total=9 passed=6 failed=0 skipped=3 —
  see test-results/recon-2.log / recon-11.log.

Risks

None — test-harness and docs only. No receiver/sender runtime change.