docs(readme): Railway deployment status 2026-04-27 — fleet 100/101/102 + P0 corpus blocker (gHashTag/trios#143)

[gHashTag]

Update README — Railway deployment status (2026-04-27)

Refs: gHashTag/trios#143 · companion comment: issuecomment-4324652513

What

Adds a new top-level section ## Railway Deployment Status (2026-04-27) immediately above ## License. The section documents:

• Live 3-seed cloud fleet trios-train-seed-100/101/102 deployed to Railway project IGLA (e4fe33bb-3b09-4842-9782-7d2dea1abc9b), env production, all SUCCESS.
• The R5-honest Gate-2 verdict: NOT DONE because the training corpus (data/tinyshakespeare.txt) is missing from ghcr.io/ghashtag/trios-trainer-igla:latest and containers fall through to a synthetic fallback yielding val_bpb=0.0000 (degenerate, not a valid R7 ledger row).
• Three pre-registered follow-up issues (P0 corpus blocker, P1 tri.rs infra mismatch, P2 stale fleet cleanup).

Why

• Anchors the public state of the cloud fleet for trios#143 to the main@ee7771f7 E2E test.
• Surfaces the P0 blocker so it can't be silently merged-around: train_loop.rs currently does silent fallback on missing corpus, which violates R5 honesty. The README now states this so the next PR (P0 fix) has clear context.

R5 / R7 / R9 status

• R5 honesty: README explicitly states 0/3 valid Gate-2 rows on this fleet, no DONE-claim.
• R7 triplet: not affected — emit logic is unchanged in this PR (docs-only).
• R9 embargo: not affected.

Falsification

• If the corpus is later baked into the image (P0 fix), this section can be amended (or kept as historical 2026-04-27 snapshot) by a follow-up PR.

Diff

• README.md: +35 lines (1 new section), no other lines touched. Docs-only PR.

Acceptance

• No code changes — docs only, CI safe.
• References issue feat(scarab): SOVEREIGN SCARAB pull-from-Postgres prototype #143 (NO-COMMIT-WITHOUT-ISSUE).
• R5 honest: explicitly says Gate-2 NOT DONE.
• Mentions follow-up issue priorities.