dd: derive the count_bytes read budget from bytes read

[relative23]

Fixes #13434.

calc_loop_bsize computed the remaining byte budget for count=N iflag=count_bytes from the bytes written so far, while below_count_limit gates the copy loop on bytes read. A conversion that writes more than it reads (conv=block record expansion, conv=sync padding of partial reads) pushes the written total past N while the read total is still below it, and the u128 subtraction underflowed: a panic in debug builds (attempt to subtract with overflow), and a wrapped huge value in release builds that unlocked a full-sized read on every following iteration — silently overshooting the count limit (the reproducer in #13434 reads 1512 input bytes instead of 1000; GNU reads exactly 1000).

Fix

Derive the remaining budget from rstat.bytes_total (bytes read), matching below_count_limit and the documented semantics of count_bytes, with a saturating subtraction so any future inconsistency degrades to a zero-sized read instead of wrapping. The now-unused WriteStat parameter is dropped.

With the fix, the #13434 reproducer reads exactly 1000 input bytes in both build profiles (like GNU). The remaining output difference vs GNU on that reproducer (92 vs 91 records) is the pre-existing chunk-local conv=block behavior, which occurs without count= as well and is out of scope here (details in the issue).

Testing

• New unit test for calc_loop_bsize (budget derived from bytes read; exhausted budget yields a zero-sized read; an overshot budget saturates to zero instead of wrapping).
• New integration test with the deterministic reproducer: panics in debug builds before the fix, passes after, and pins the input side (1+1 records in, i.e. exactly 1000 bytes read) as well as the expected output size.
• cargo test -p uu_dd (89 tests) and the dd integration suite (123 tests) pass.
• Outputs are byte-identical to main across non-expanding count configurations (blocks, plain count_bytes, count_bytes with conv=sync on full reads, split ibs/obs).

[codspeed-hq]

Merging this PR will not alter performance

✅ 339 untouched benchmarks
⏩ 46 skipped benchmarks¹

---

_{Comparing relative23:dd-count-bytes-loop-bsize (0a07694) with main (7807c8d)}

Footnotes

1. 46 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[github-actions]

GNU testsuite comparison:

Skip an intermittent issue tests/pr/bounded-memory (fails in this run but passes in the 'main' branch)
Skip an intermittent issue tests/tail/overlay-headers (fails in this run but passes in the 'main' branch)
Skip an intermittent issue tests/tail/pid-pipe (fails in this run but passes in the 'main' branch)
Note: The gnu test tests/tail/tail-n0f is now being skipped but was previously passing.
Congrats! The gnu test tests/expand/bounded-memory is now passing!
Congrats! The gnu test tests/rm/many-dir-entries-vs-OOM is now passing!

[sylvestre]

this is a long comment, please make it shorter

[relative23]

Done — shortened it (and the matching test comment) to two lines; the full rationale stays in the commit message.