fastq: Add RecordIndex and IndexedReader for O(1) random access

RecordIndex builds an offset index in a single scan, replacing the
separate count_lines, build_record_index, and estimate_avg_record_size
functions. IndexedReader wraps a seekable reader + index and provides:
  - read_record_at(i): seek to known offset, no boundary scanning
  - read_records_at(indices): batch sorted reads
  - skip_ahead_sample(target, rng): geometric jumps in record space

This eliminates the byte-level boundary detection (find_record_after)
that was unreliable for paired-end data. Skip-ahead now jumps between
index entries instead of byte positions.

Removed from subsample.rs: skip_ahead_sample_file, build_record_index,
read_record_at, find_record_after, exponential_sample,
estimate_avg_record_size, count_records_from_index.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: jdidion