…e to 2.0.72.Final

### What changes were proposed in this pull request?

This PR aims to Upgrade Netty to 4.1.122.Final and Netty-tcnative to 2.0.72.Final.

### Why are the changes needed?

To bring the Java 24 related update and the latest bug fixes.
> We decided to enable sun.misc.Unsafe by default again when using JDK24+ (when possible) as disabling it by default on this java versions did come with quite a performance regression.

- https://netty.io/news/2025/06/03/4-1-122-Final.html
    - netty/netty#15296
    - netty/netty#15248
    - netty/netty#15122

### Does this PR introduce _any_ user-facing change?

No.

### How was this patch tested?

Pass the CIs.

### Was this patch authored or co-authored using generative AI tooling?

No.

Closes #51145 from dongjoon-hyun/SPARK-52438.

Authored-by: Dongjoon Hyun <dongjoon@apache.org>
Signed-off-by: yangjie01 <yangjie01@baidu.com>

…rect buffers (#298)

## Summary

Netty 4.1.120 and 4.1.121 disabled their internal `sun.misc.Unsafe`
usage by default
([netty/netty#14975](netty/netty#14975)) to
silence JEP 498 warnings on JDK 24+. With Unsafe disabled,
`PlatformDependent.useDirectBufferNoCleaner()` returns `false`: every
Netty direct buffer (pooled **and** unpooled) falls back to
`ByteBuffer.allocateDirect` → `java.nio.Bits.reserveMemory`, which is
bounded by `-XX:MaxDirectMemorySize`. The result is `OutOfMemoryError`
under heavy ingestion even though `-Dio.netty.maxDirectMemory` was set
generously.

Netty 4.1.122
([netty/netty#15296](netty/netty#15296)) reverts
the default. Until HerdDB upgrades (commit c0cf458 downgraded us
4.1.130 → 4.1.121 for issue #274), add
`--sun-misc-unsafe-memory-access=allow` to:

- `herddb-services/src/main/resources/bin/setenv.sh` — applies to every
JVM launched via `bin/service` (server, indexing-service, file-server,
bookkeeper) and to the CLI tools that inherit `JDK_JAVA_OPTIONS`
(`herddb-cli.sh`, `vector-bench.sh`, `indexing-admin.sh`).
- `pom.xml` — both Surefire `argLine`s (default and `ci` profile) so
unit tests on JDK 25 don't take the slow `Bits.reserveMemory` path
either.

## Verification

Reproduced locally on Eclipse Temurin **JDK 25.0.2** with the in-tree
Netty **4.1.121**.

**Before** (no flag): server saturates `-XX:MaxDirectMemorySize=1g`,
OOMs at 160k/1M rows even with `-Dio.netty.maxDirectMemory=4g`.
`netty_used_direct_memory_bytes` Prometheus gauge stuck at `-1` (counter
inactive).

**After** (with flag): server peak direct memory ~300 MiB, **0
OutOfMemoryErrors** through the full gist1m bench (1M vectors × 960-dim,
8 ingest threads, 5000 batch, post-ingest checkpoint, recall queries).
`netty_used_direct_memory_bytes` reports actual usage (counter active).

| Metric | Before | After |
|---|---|---|
| `netty_used_direct_memory_bytes` | `-1` | active (~134 MiB at end) |
| Server direct mem peak | hits 1024/1024 MiB cap | ~300 MiB / 4 GiB |
| OOMs in 1M-row ingest | 30–66 | **0** |
| Ingest throughput | bench fails | 20,332 rows/s |
| Recall@10 | bench fails | 0.91 |

## Also included

`herddb-services/examples/local-bench/install.sh` got a small batch of
bench-tuning changes that emerged from this investigation:

- `-Dherddb.file.requirefsync=false` for both server and
indexing-service (benchmark-only — never use this in production).
- Explicit `-XX:MaxDirectMemorySize` and `-Dio.netty.maxDirectMemory`
for both services (was leaving Netty at `=0`/JVM-bound; now explicit and
consistent).
- Indexing-service heap reduced from 40g → 30g so the `(server) + (IS)`
budget fits in a 61 GiB host with headroom for the OS.
- Documentation comments explaining the layout.

These are isolated to the `examples/local-bench` directory and don't
affect production defaults.

## Test plan

- [ ] CI runs all unit tests with the new Surefire `argLine` on JDK 25
- [ ] Smoke-test the `local-bench` example: `./teardown.sh &&
./install.sh && ./scripts/run-bench.sh --dataset gist1m -n 1000000
--ingest-threads 8 --batch-size 5000 --checkpoint --wait-for-indexes`

## Long-term follow-up

When HerdDB is upgraded back to **Netty 4.1.122+** (or **4.2.x**, which
is FFM-based and Unsafe-free), the
`--sun-misc-unsafe-memory-access=allow` flag in `setenv.sh` becomes
redundant and the comment block can be removed.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>