The branch was guarded against len(slab) < pos+n, but both NewArena
and the grow path use make([]byte, n) which always returns len==cap.
Remove the dead branch and tighten the comment to document why the
3-arg slice is always in bounds.

Also clarify ResetAndShrink doc for the maxKeep <= 0 case and add a
test proving Reset reuses the same backing array.

Decodes Output using a caller-provided Arena for the locking-script
[]byte. Existing ReadFrom is unchanged. Rejects script lengths greater
than MaxArenaAlloc (1 GiB) to prevent unbounded allocation from
adversarial input.

Mirrors Input.ReadFrom / ReadFromExtended via an internal helper
readFromWithArena. Pulls the unlocking-script byte slice (and the
extended-format PreviousTxScript) from a caller-provided Arena.
Existing ReadFrom / ReadFromExtended unchanged. Fixed-size fields
(previousTxID, prevIndex, sequence, prevSatoshis) use stack arrays.
Rejects script lengths exceeding MaxArenaAlloc.

a.Alloc(0) on a fresh arena returns a nil slice, breaking equivalence
with the existing ReadFrom which uses make([]byte, 0) (non-nil empty).
Add a sentinel branch and a zero-length-script fixture. Same pattern
as Input.ReadFromWithArena.

Mirrors Tx.ReadFrom field-for-field with extended-format detection
preserved. Routes per-script allocations through the caller-provided
Arena via Input/Output ReadFromWithArena variants. Existing ReadFrom
is unchanged. Verified equivalent on canonical tx fixtures and a
320 MB OP_RETURN payload.

Lets callers (e.g. teranode's catch-up stream readers) reuse a single
scratch buffer across millions of txid hashes instead of allocating
tx.Bytes() per call. Honours the existing SetTxHash cache: when cached,
the scratch buffer is passed through unchanged and no allocation occurs.

Rejects script lengths greater than MaxArenaAlloc (1 GiB) in the
existing Output.ReadFrom and Input.readFrom (unlocking + extended
PreviousTxScript). Previously a malformed or adversarial varint
claiming e.g. 4 GiB would cause an immediate make([]byte, 2^32)
allocation and likely OOM. The cap remains loose enough to admit
legitimate large BSV transactions (~320 MB).

Compares Tx.ReadFrom vs Tx.ReadFromWithArena for single-tx and
5000-tx subtree-style streams. Compares HashTxIDInto vs the
existing TxIDChainHash.

Benchmark results (Apple M3 Max, go test -benchmem -count=1):

BenchmarkTx_ReadFrom_vs_WithArena/ReadFrom          1682888   702.2 ns/op  1096 B/op  44 allocs/op
BenchmarkTx_ReadFrom_vs_WithArena/ReadFromWithArena 1948824   653.5 ns/op   696 B/op  39 allocs/op
  → -7% ns/op, -36% B/op, -11% allocs/op per tx

BenchmarkSubtreeDecode_5kTxs/PerTxReadFrom    344   3407017 ns/op  5240099 B/op  215001 allocs/op
BenchmarkSubtreeDecode_5kTxs/PerSubtreeArena  402   2982810 ns/op  3247887 B/op  190001 allocs/op
  → -12% ns/op, -38% B/op, -12% allocs/op per 5k-tx batch

BenchmarkHashTxIDInto_vs_TxIDChainHash/TxIDChainHash  3291541   353.3 ns/op  608 B/op  2 allocs/op
BenchmarkHashTxIDInto_vs_TxIDChainHash/HashTxIDInto   3892976   298.6 ns/op    0 B/op  0 allocs/op
  → -15% ns/op, -100% B/op, zero allocs

Pulls the varint + cap + alloc-or-make + ReadFull pattern out of
Output.ReadFromWithArena and the two script-reading blocks in
Input.readFromWithArena into a single readArenaScript helper.

Net effect:
- ~60 lines of duplicated decode logic removed
- Single source of truth for the MaxArenaAlloc cap message and the
  nil-arena fallback path
- Error message labels are now consistent (lockingScript /
  unlockingScript / prevTxScript) across both cap and short-read paths

Addresses SonarQube 3.2% duplication-on-new-code (gate is <= 3%).

- Collapse TestTx_Clone_IsolatesFromArena + TestTx_ShallowClone_IsolatesFromArena
  into a single table-driven TestTx_CloneVariants_IsolateFromArena.
  Eliminates ~80% of the duplicated test body that pushed SonarQube
  duplication-on-new-code to 8.4%.
- tx_arena_test.go:98: serialisation -> serialization (misspell).
- arena_test.go: TestArena_Alloc_ZeroOnZeroValueArena uses require.Empty
  per testifylint, not require.Len(..., 0).

- serialisation -> serialization (misspell, 2 occurrences)
- Drop "Note:" prefix on the SetTxHash-cache clarification (godox
  flags it as a TODO-style marker). Same content, no prefix.