…t-circuit

Previously, when engine_newPayload was called twice for the same bad
block, the second call short-circuited via the badHeaders LRU with a
generic "previously known bad block" string in ValidationError —
discarding the specific reason (e.g. "max initcode size exceeded") that
the first call would have returned.

This was harmless for production until the hive warm-daemon client pool
(PR #20812) started reusing erigon processes across EEST tests. The pool
calls debug_setHead(0) between tests but does not clear in-memory
caches, so a later test that happens to send a payload with a
block-hash also produced by an earlier test gets the cached
short-circuit instead of fresh validation — and EEST's
ErigonExceptionMapper has no rule for the generic string, so the test
fails with the wrong exception even though the block was correctly
rejected on first sight.

Cache the original validationErr string alongside the lastValidHash and
replay it on hit. Fall back to "previously known bad block" when the
cache entry was produced by a path that didn't surface a validation
string (sync-time downloader, "invalid block number" header check). For
parent-inheritance hits, wrap as "ancestor 0x… rejected: <parent err>"
so the cause is still traceable when the child hash hadn't been seen
before.

Beyond the EEST flake, this is also a better Engine API experience for
production: a CL that resubmits a bad payload (retry-after-glitch,
debugging session) now sees the same actionable error twice instead of
a generic short-circuit on retry.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

- Drop the BadHeaderEntry doc comment (the field names are
  self-descriptive and the prior wording incorrectly listed the
  "invalid block number" header check as an empty-string path —
  this PR populates it). Per #21355.
- Shrink the badHeaders LRU from 10_000 to 96. Each entry now holds a
  heap-allocated string that won't be GC'd while cached, and the
  realistic working set on a healthy node is tiny.
- Add unit tests pinning the ReportBadHeader/IsBadHeader round-trip
  for the validation-error string, including the empty-err fallback
  and the later-report-overrides path.
- Drop the scenario-narrative comment block above the IsBadHeader
  short-circuit per #21355.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…cache-validation-err

Adds a small generic helper

    func Deref[T any](p *T) T

and adopts it in six spots that all hand-rolled the same shape — an
explicit nil check followed by *p, with the zero value of T as the
implicit fallback. Five different pointer element types are
consolidated (*string, *[]byte, *uint64, *log.Logger, plus the
package-private derefString that this PR introduced earlier).

Also trims the glamsterdam-devnet workflow comment: now that the
EIP-7954 test_max_initcode_size flake is fixed by this PR, the
budget breakdown no longer needs to enumerate the "Mode A / Mode B"
framing — the one remaining flake slot is for #21364.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>