Fix safe_dist under-reporting for multi-block inputs

[meeq]

apultra_write_block tracked nCurSafeDist as (i - nStartOffset) - nOutOffset: a per-block local delta between input consumption and output emission. Because nOutOffset resets to 0 at the top of each apultra_write_block call (pOutData is already advanced by the caller), and nStartOffset is the current block's offset in the input window, this measure ignores the accumulated decoder-output-vs-decoder-input lead from previous blocks.

For single-block inputs (< 1 MiB) the local delta equals the global delta and the bug is invisible. For multi-block inputs the encoder writes a safe_dist that is too small, and any consumer doing in-place decompression (allocating bufsize from safe_dist) sees its output pointer overtake the unread compressed input and corrupt subsequent reads.

Fix: thread nPriorDelta (the cumulative decoder-output - decoder-input at block start) through apultra_compressor_shrink_block → apultra_optimize_and_write_block → apultra_write_block, and add it to nCurSafeDist. The top-level loop in apultra_compress already has the cumulative nOriginalSize and nCompressedSize, so nPriorDelta is (nOriginalSize - nDictionarySize) - nCompressedSize. Dictionary bytes aren't emitted by the decoder, so they're excluded from the input side.

The compressed bitstream content is unchanged by this fix; only the reported safe_dist (and downstream in-place margin) changes.