VarHandle multi-byte reads for ByteString lastIndexOf

[pjfanning]

• Relates to Optimise ByteString lastIndexOf #2150 (there is non SWAR PR too that may work out better)

SWARUtil — extended VarHandle infrastructure

• Use byteArrayViewVarHandle instances for short (BE+LE), int (BE+LE), long (LE)
• Added getLastIndex(word) — finds rightmost byte match in SWAR result using numberOfTrailingZeros

ByteString

• Use SWARUtil for multi byte processing

Significant improvement in benchmark.

With Changes
[info] Benchmark                                                              Mode  Cnt          Score           Error  Units
[info] ByteString_lastIndexOf_Benchmark.bs1_lastIndexOf                      thrpt    3  107880125.372 ± 247014962.175  ops/s
[info] ByteString_lastIndexOf_Benchmark.bs1_lastIndexOf_byte                 thrpt    3   88478677.724 ± 181685386.699  ops/s
[info] ByteString_lastIndexOf_Benchmark.bss_lastIndexOf_best_case            thrpt    3   58313896.505 ± 806477112.611  ops/s
[info] ByteString_lastIndexOf_Benchmark.bss_lastIndexOf_far_index_case       thrpt    3   17472854.418 ±  94251978.262  ops/s
[info] ByteString_lastIndexOf_Benchmark.bss_lastIndexOf_far_index_case_byte  thrpt    3   19332603.332 ±   2245521.678  ops/s
[info] ByteString_lastIndexOf_Benchmark.bss_lastIndexOf_worst_case           thrpt    3   20053173.893 ±   2385070.269  ops/s

Without Changes
[info] Benchmark                                                              Mode  Cnt          Score          Error  Units
[info] ByteString_lastIndexOf_Benchmark.bs1_lastIndexOf                      thrpt    3   33298670.109 ± 63631788.200  ops/s
[info] ByteString_lastIndexOf_Benchmark.bs1_lastIndexOf_byte                 thrpt    3   27021365.677 ±  5557597.483  ops/s
[info] ByteString_lastIndexOf_Benchmark.bss_lastIndexOf_best_case            thrpt    3   34686168.040 ±  7080975.674  ops/s
[info] ByteString_lastIndexOf_Benchmark.bss_lastIndexOf_far_index_case       thrpt    3     635863.919 ±    69057.413  ops/s
[info] ByteString_lastIndexOf_Benchmark.bss_lastIndexOf_far_index_case_byte  thrpt    3     597510.828 ±    29481.618  ops/s
[info] ByteString_lastIndexOf_Benchmark.bss_lastIndexOf_worst_case           thrpt    3    2125744.038 ±   196176.723  ops/s

[He-Pin]

Good performance work, but a few concerns:

Concerns:

1. ByteString1C.lastIndexOf and ByteString1.lastIndexOf -- The SWAR-based specialized lastIndexOf(elem: Byte, end: Int) looks correct but the unrolledLastIndexOf tail handling is complex. The tests cover basic cases but I'd recommend adding a test that specifically targets the boundary between tail and full 8-byte chunks (e.g. length=15, element at position 14).

2. SWARUtil.getLastIndex -- the doc says Currently only supports big endian. However it's used by the lastIndexOf implementation which reads via SWARUtil.getLong(bytes, chunkStart) (the default big-endian path). This is correct, but worth explicitly documenting to prevent future misuse.

3. The PR description says WIP and needs benchmarking. I'd recommend not merging until benchmarks confirm the SWAR lastIndexOf is actually faster than the simple loop for typical ByteString sizes (many ByteStrings are small, where SWAR overhead may not pay off).

4. ByteIterator overrides -- using byteOrder == ByteOrder.BIG_ENDIAN boolean flag is cleaner than the if/else chain in #2847. Consider whether these two PRs should be coordinated to avoid duplicate implementations.

[pjfanning]

I'm breaking this up into smaller PRs

[pjfanning]

I removed the parts of the original PR and focused on SWAR lastIndexOf. The other parts were merged already in other PRs.