Add VTGate sequence range caching

[ramnov]

What's this?

Adds an opt-in, in-memory VTGate cache for VSchema auto_increment sequences. The goal is to avoid a sequence-primary RPC on every insert, especially when VTGate and the sequence primary are in different regions.

How it works

• Reserves ID blocks through the existing NEXT N sequence path; VTTablet is unchanged.
• Keeps an independent, concurrency-safe cache per VTGate process and sequence.
• Serves cache hits locally and reserves a new contiguous block when the current range cannot satisfy an insert.
• Adds --enable-sequence-cache and --sequence-cache-size flags plus per-sequence metrics.
• Leaves background prefetch and reset/migration invalidation out of this initial proof of concept.

Testing

• Sequence cache unit tests, including concurrent allocation and refill failures
• Linux race detector
• VTGate enabled/disabled sequence integration tests
• VTGate engine insert tests
• Size generator verification