…unknown (netty#16207)

Motivation:
When the JVM cannot report UNALIGNED support (e.g., on some ARM
platforms or in restricted environments), Netty falls back to an
architecture-based guess. If that guess yields UNALIGNED = false,
multi-byte reads (getLong,
getInt, getShort) degrade to byte-by-byte access (8× getByte) even
though the JIT could still optimize at runtime. There is currently no
distinction between "the platform is known to not support unaligned
access" and "we
simply don't know."
Modification:
Introduce UNALIGNED_AVAILABLE flag to distinguish whether UNALIGNED info
came from the runtime (true) or from an architecture-based guess
(false). When unaligned info is unavailable, use VarHandle instead of 8×
getByte so
HotSpot can emit the optimal access strategy at runtime.
- PlatformDependent0: add UNALIGNED_AVAILABLE flag; add
io.netty.unalignedAccess=true|false|unavailable system property.
- PlatformDependent: expose isUnalignedAvailable(); allow VarHandle
initialization even when Unsafe is available.
- UnsafeByteBufUtil: use `VarHandle` for multi-byte `byte[]` operations
(`getShort/getInt/getLong`, `setShort/setInt/setLong` and their LE
variants) when unaligned info is unavailable.
- UnpooledUnsafeDirectByteBuf: use `VarHandle` for multi-byte
`ByteBuffer` operations (`_getShort/_getInt/_getLong`,
`_setShort/_setInt/_setLong` and their LE variants) when unaligned info
is unavailable.
- ByteBufUtil: enable SWAR indexOf/lastIndexOf via VarHandle when
unaligned info is unavailable.
Three-tier access strategy:

| Condition | Strategy | Path |
|---------|---------|------|
| UNALIGNED = true (known supported) | Unsafe direct access | single
`Unsafe.getLong()` |
| UNALIGNED info unavailable (unknown) | VarHandle | JIT chooses optimal
access |
| UNALIGNED = false (known unsupported) | 8× getByte | safe byte-wise
assembly |

---
Result:
Fixes netty#15781.
Benchmark (JDK 24.0.1, Apple Silicon, JMH 1.36):
### Benchmark Summary

| Benchmark | false (8× getByte) | unavailable (VarHandle) | Improvement
|

|----------|--------------------|--------------------------|-------------|
| getLongDirect | 2.144 ns/op | 1.764 ns/op | 17.7% faster |
| getLongHeap | 2.152 ns/op | 1.119 ns/op | 48.0% faster |

Sorry for the delay in submitting this PR.
If anything looks incorrect or needs adjustment, I will update it
promptly.
Thanks for the review.

---------

Co-authored-by: Norman Maurer <norman_maurer@apple.com>
Co-authored-by: Chris Vest <christianvest_hansen@apple.com>

(cherry picked from commit 0ee9723)