b23da1a changed getStateObject to (a) seed refreshVersionedAccount
with UnknownVersion and (b) drop the synthetic accountRead AddressPath
recording, to quell validator path==AddressPath false positives behind
the SD-recreate flake fix.

Both are now redundant: the merged main carries #21286 (exclude own-tx
writes from the SD revival check), which independently covers the
+1602 / CREATE2-collision flake mode, and the validator already keys
on the precise IncarnationPath signal. They are also harmful — without
the accountRead recording the account is missing from the parallel-
built block access list, so selfdestruct/create-destruct family blocks
fail "block access list mismatch" under the parallel runner (serial
unaffected).

Restoring the original recording clears ~102 EEST bal-devnet
(Amsterdam) failures. TestDeleteRecreateSlotsAcrossManyBlocks stays
200/200 under GOMAXPROCS=2 (100 with -race) — the flake fix is carried
by #21286 and b23da1a's validator/scheduler changes.

…tives (erigontech#21294)

## Summary

`TestDeleteRecreateSlotsAcrossManyBlocks` flaked ~3–5% under
`GOMAXPROCS=2 -race` with `ERIGON_EXEC3_PARALLEL=true`, in two modes: a
+1602 gas mismatch (a CREATE2 collision burned inner gas on a stale
CodeHash read) and a runaway validator-invalid retry loop. Several
distinct validator/recording false positives contributed; this PR fixes
them in turn.

- **`getStateObject`** initialised `accountVersion` to `sdb.Version()`
(the worker's own incarnation), gating out every prior-tx sub-read
update in `refreshVersionedAccount` and pinning source/version to a
value vm never wrote. Switched to `UnknownVersion`.
- **`accountRead`** synthesized an AddressPath read using the version
that flowed through `refreshVersionedAccount` — usually a sub-read's
(Balance/Nonce/etc) version. The validator's `path==AddressPath` checks
then fired against an entry vm never wrote. Removed the synthetic
recording; sub-reads are already recorded by `versionedRead`.
- **`versionedRead` MVReadResultNone**: exclude AddressPath from the
`readStorage==nil` probe recording. `getStateObject` probes vm then
falls back to the stateReader (account does exist); the "we read
nothing" recording made the validator's path==Address branch invalidate
any pre-existing account that a prior tx in the block wrote BalancePath
for.
- **`validateRead`** wrapped in `validateReadImpl` with a `recursive`
flag. Two false-positive paths in the cross-validate recursion produce
spurious Invalid when invoked as a synthetic probe (readVal nil, source
borrowed): the path==AddressPath BalancePath-presence check, and the
`source!=MapRead+nil-readVal` tiebreaker. Both gated on `!recursive`.
- **`validateRead`** also gained an SD-staleness cross-check in the
`MVReadResultDone` branch: a later tx self-destructing without revival
means the read predates the destruct and the serial path returns zero
via the SD-zero short-circuit; `checkVersion` alone misses this because
SD doesn't write the read's own path. `revivalLimit = txIndex - 1`
excludes our own writes (validation runs post-flush).
- **Deferred queue**: on ErrDependency with no effective blocker, OR
validator-invalid, push to a new `deferred[]` queue instead of
`pending`. `scheduleExecution` drains a deferred tx only when its
predecessor has validated AND no worker at a lower index is in flight.
Lower-indexed workers' flushes land at indices visible to N's reads via
`vm.Read`'s `floor(N-1)`; higher-indexed ones don't. Non-deferred txs
keep dispatching speculatively. A safety net force-drains when pending
is empty and no workers are in flight.
- **`tooManyRetries`** helper + cap on validator-invalid retries
(mirrors the worker-error path) so a genuine loop can't spin forever.

erigontech#21286 (yperbasis) fixes the same `revivalLimit = txIndex - 1` insight
in the runtime SD short-circuit (`versionedRead` +
`versionedStateReader.ReadAccountData`); this PR fixes the
validator/recording side. The two are complementary — the runtime fix
prevents the bad read at source, this PR prevents the validator from
generating false-positive invalidations on legitimate reads.

## Test plan

- [x] `make lint` clean (run multiple times).
- [x] 200/200 pass on `TestDeleteRecreateSlotsAcrossManyBlocks` under
`GOMAXPROCS=2 ERIGON_EXEC3_PARALLEL=true go test -race` (vs ~95-97/100
baseline).
- [x] 100/100 pass on the same test rebased onto a clean `origin/main`.
- [x] No EEST regression: `cancun` blocktests (8389) + `eip4844_blobs`
race-build (6991) + the previously-affected
`test_sufficient_balance_blob_tx_pre_fund_tx` (864) all pass under the
12-worker parallel runner.
- [ ] CI green on race-tests + EEST parallel shards.