The parallel frame reader's metadata-direct path wrote each data file's
rows into shared per-column arrays at a fixed offset without bounding the
row count, so a table whose per-file numRecords statistic under-counts the
actual rows (possible for externally written Delta tables) could overrun
its slice into the next file's region under concurrent writes.

- Add the per-file row-count overflow guard in FrameReaderDeltaParallel
  .readDirect, matching the matrix reader: fail fast with a clear message
  instead of risking overlapping concurrent writes or an array overrun
- Reuse DeltaKernelUtils.typeCode/T_* in FrameReaderDelta instead of a
  forked R_* table and instanceof cascade, keeping the frame and matrix
  type dispatch in lockstep; drop the now-unused type imports
- Extract awaitFileTasks in FrameReaderDeltaParallel to share the pool
  lifecycle across both read paths and restore the interrupt flag when a
  parallel read is cancelled
- Add a unit test covering the adaptive target-file-size flag on/off and
  the floor/cap clamp boundaries
- Clarify the adaptive-size javadoc floor wording, the createWriteEngine
  batch-size comment, and rename opaque locals (names2, bcs/bss)

The single-threaded frame reader extracted each batch into temporary
per-batch arrays and then concatenated them into the final column arrays,
allocating and copying every column twice. Delta's per-file numRecords
statistic already provides exact row counts from metadata, so the output
can be sized up front and decoded in a single pass with no intermediate
buffers.

- Rewrite FrameReaderDelta.readFrameFromHDFS to pre-size one typed array
  per column from the metadata row counts and decode each data file
  straight into its row offset (with the same per-file overflow guard as
  the parallel reader); fall back to the buffered extract-then-concatenate
  path only when exact counts are unavailable (missing statistics or
  deletion vectors present)
- Move allocColumn/createColumn/extractColumnInto up to FrameReaderDelta so
  the serial and parallel readers share one copy instead of duplicating the
  per-type column dispatch
- The parallel reader's single-file/low-thread fallback now also decodes in
  a single direct pass

Route the Delta frame readers' column allocation and wrapping through
ArrayFactory instead of reader-local per-type switches:

- Add ArrayFactory.allocateBacking(ValueType,int) as the single
  ValueType -> raw backing array mapping (the inverse of
  create(ValueType,Object)), and remove the duplicate allocColumn /
  createColumn / extractColumn / buildColumn switches from
  FrameReaderDelta and FrameReaderDeltaParallel. The buffered fallback
  now reuses the same alloc + extractColumnInto + concatColumn
  primitives as the direct path.
- Make create(ValueType,Object) bit-pack boolean columns above the
  switch point into a BitSetArray (mirroring allocateBoolean), so Delta
  reads produce the same compact representation as every other frame
  reader instead of a byte-backed BooleanArray.
- Simplify allocate(ValueType,int) to compose create + allocateBacking,
  keeping only the boolean special case (empty BitSet backing) and
  moving the UINT4/UINT8 fallback warning into allocateBacking.
- Move useDirectPath to FrameReaderDelta so both readers share it.

Add timing/throughput benchmarks for the serial and parallel Delta
frame readers:

- DeltaFrameRead in the performance suite (dispatched as Main id 18):
  writes a random frame to a temp Delta table once as untimed setup,
  then repeatedly reads it back under timing for serial/parallel/both
  modes with a configurable writer target file size. Suitable for
  running under async-profiler.
- DeltaFrameReadPerf: JUnit-based manual micro-benchmarks (all @ignore
  so they never run in the normal build) covering direct-vs-buffered
  serial reads, adaptive file sizing, target-size and batch-size sweeps,
  and schema-composition breakdowns.

Reformat the new Delta frame reader/writer and test files with the
project Eclipse style (dev/CodeStyle_eclipse.xml). Whitespace and layout
only, no behavioral changes.

- Replace the bespoke genMixedFrame generator in DeltaFrameReadWriteTest
  with TestUtils.generateRandomFrameBlock, removing duplicated random
  frame generation code.
- Add DeltaFrameSparkInteropTest exercising cross-engine round-trips
  against the reference Delta Spark connector: SystemDS-written multi-file
  frame read by Spark, Spark-written multi-file frame read by the serial
  and parallel SystemDS readers, and a Spark table with deletion vectors
  read via the buffered selection-mask path. Comparisons are keyed by a
  unique id column so no row order is assumed.

…elUtils

The prior commit ran the Eclipse formatter over the entire files, reflowing
many lines this PR does not touch. Restore both files to a minimal diff that
carries only the intended changes: the null-result check moved above the
metadata refresh in FrameObject, and the debug log for the adaptive writer
file-size decision in DeltaKernelUtils.