Make Java-backed read coalescing configurable #11366

Java-backed Vortex readers cannot configure IO coalescing. JavaReadable::coalesce_config() unconditionally returns CoalesceConfig::object_storage(): a 1 MiB gap and 16 MiB maximum merged span on upstream develop at 2e1b05b84.

Different storage systems and projections need different tradeoffs between request count and extra bytes read. A caller supplying NativeReadable, including Spark's Hadoop-backed readable, currently cannot make that choice.

Expose default coalesceDistance() and coalesceMaxSize() methods on NativeReadable, capture and validate them once when registering a source, and expose matching Spark read options. Preserve upstream's existing 1 MiB/16 MiB defaults. Validate non-negative distance and positive maximum, with alignment headroom for Java's ByteBuffer capacity limit.

QC issue #94 and QC PR #95 provide an implementation. QC uses a different default gap (64 KiB); the upstream port must retain 1 MiB. The proposed DataFusion-specific local IO policy in #11338 does not expose configuration to Java callers.

Splitting individual reads larger than ByteBuffer's capacity is a separate follow-up; this change only makes the existing coalescing limits configurable.