docs(VALIDATION): record the Postgres-tier chaos acceptance — attributable resume-from-partial proven

Restructures §4 into two tiers and records the new evidence:

- §4 intro + §4.1: Tier 0 (PGlite, byte-identity, ACCEPTED 2026-07-04) is
  kept verbatim as the record of what it alone proves (atomicity,
  idempotence, byte-identical completion), with a bridge noting the
  resume-from-partial gap it scopes out is now closed by Tier 1.
- §4.2 (new): Tier 1 (native Postgres, crash-durable, ACCEPTED
  2026-07-05). Carries the acceptance numbers (203 kills / 27
  completed-verified / 155 kills at partial coverage / 17 completions
  from partial persisted state; status pass), the staircase
  parameterization (PORTAL_CHUNK_BLOCKS=2000, fixed, no read-ahead), the
  logical-digest identity definition (surrogate-serial-id excluded,
  natural-key total order), and the candor stories:
    * run 2 — kept as a fail (driver leaked a second concurrent writer →
      factory_addresses ×2); its kills excluded from the totals; the
      store-level lesson filed #53, write-side fix merged #54.
    * run 20 — a digest false-FAIL reclassified to pass: a non-
      transactional serial SEQUENCE advanced across a rolled-back flush,
      so a correct resume re-flushed identical content at shifted ids; the
      digest now excludes surrogate serials (methodology finding).
- §5.4: adds #52 (PGlite WAL tear — harness-backend finding), #53
  (factory_addresses idempotence, write-side fix merged #54), and the
  run-20 surrogate-id digest methodology row; updates the "queued
  re-parameterized campaign" paragraph → it ran and closed the gap (→§4.2).
- §6: partial-resume is now among what a reader can rely on today.
- §7: repro table row + env-var contract for the pg-tier tools.

Relative dates converted to absolute; issues linked as the file does.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: dzhelezov