Hash64: hybrid inline/getBytes UTF-8 path in updateString

[brharrington]

Replaces the previous single-path updateString with a length-based hybrid:

• For length ≤ 128 (typical metric tag values) or any non-String CharSequence: inline UTF-8 char loop, no allocation, no per-call encoder setup overhead.
• For longer String inputs: fall back to getBytes(UTF_8) + Unsafe long-stride updateBytes, where the intrinsified UTF-8 encoder and 8-byte-per-iter writes win on throughput.

Output is bit-identical to updateBytes(s.getBytes(UTF_8)) regardless of which path is taken; the existing Hash64Test.checkString assertions cover both sides of the threshold (random strings of length 0-999).

Other changes:

• New private writeByte helper handles per-byte stripe writes and rollover; updateByte delegates to it to avoid duplicate logic.
• build.gradle: pin the jmh task to JDK 25 so benchmarks reflect the modern JIT, independent of the JDK 8 source toolchain.

JMH on JDK 25 (M1; see IdHash.java javadoc for the full table):

  IdHash.hash64             1.25M -> 2.05M ops/s   (+64%)
  IdHash.shortAsciiString     20M -> 22M           (+13%)
  IdHash.unicodeString        12M -> 14M           (+19%)
  IdHash.longAsciiString    4.15M -> 4.20M         (+1%)
  IdHash.hash64_2           2.98M -> 2.73M         (-9%)