iter6: revert depth bump + WRITEBACK_CHUNK_BYTES 32->8 MiB

iter5 (256-frame channel) regressed -1.8 MB/s vs iter4. Reverting to
32 frames.

iter4 sample pattern shows clear ~30 s oscillation (peak 45 MB/s →
dip 3 MB/s → recovery). Matches Linux vm.dirty_expire_centisecs
default (30 s). 32 MiB chunks at 25 MB/s issue WAIT_AFTER every
~1.3 s, which can't outrun the kernel's own page-age limit, so
pages buildup then flush in bursts. Smaller 8 MiB chunks
(WAIT_AFTER every ~0.33 s) should keep the dirty-page set young
and eliminate the periodic flush-burst dip.

src/io/pipeline.rs

@@ -90,16 +90,7 @@ pub const DEFAULT_PIPELINE_DEPTH: usize = 4;
 /// Read pipeline depth. Larger buffer compensates for drive variability
 /// and NFS sync_file_range stalls; keeps ISO reader thread fed even when
 /// consumer blocks on write.
-///
+pub const READ_PIPELINE_DEPTH: usize = 32;
-/// iter5 (2026-05-17): bumped 32 → 256 frames. autorip iter4 measured
-/// dips to 2.8 MB/s with the previous 32-frame channel (~1.6 MiB at
-/// ~50 KB/frame avg). When the producer thread hits any micro-stall
-/// (UDF metadata cache miss, NFS RTT, decrypt key lookup), a 1.6 MiB
-/// buffer drains in <100 ms and the consumer sits idle. 256 frames is
-/// ~12 MiB — ~3-4 seconds of consumer drain at 4 MB/s worst-case
-/// sustained output rate, enough to coast through any single-event
-/// producer pause without starving the consumer.
-pub const READ_PIPELINE_DEPTH: usize = 256;
 
 /// Write pipeline depth. Smaller buffer reduces backpressure risk when
 /// sync_file_range blocks; prevents producer from accumulating too much

src/io/writeback_file/mod.rs

@@ -124,7 +124,13 @@ use super::writeback::WritebackPipeline;
 /// Granularity at which the Linux writeback pipeline issues
 /// `sync_file_range` / `posix_fadvise(DONTNEED)` pairs. 32 MiB is the
 /// historical default — bounded-cache pressure stays at ~2 × this size.
-const WRITEBACK_CHUNK_BYTES: u64 = 32 * 1024 * 1024;
+///
+/// iter6 (2026-05-17): 32 → 8 MiB. iter4 data showed ~30 s
+/// oscillation (peak 45 → dip 3 MB/s with ~30 s period) matching
+/// Linux's `vm.dirty_expire_centisecs` (30 s default). Smaller chunks
+/// keep dirty pages younger so the kernel flusher daemon's bursts are
+/// shorter.
+const WRITEBACK_CHUNK_BYTES: u64 = 8 * 1024 * 1024;
 
 /// Maximum bytes outstanding in the muxer → writer-thread ring. Sized
 /// to cover ~4 s of muxer output at a 32 MB/s peak — enough to absorb a