Fix MAT v5 reader: miCOMPRESSED elements are not 8-byte padded

[jyyen]

Fixes #12.

Problem

MatData::read fails on real StaMPS/scipy .mat files with errors like small data element at byte 2240 has 55552 bytes. The parser already handles zlib (miCOMPRESSED) and small elements, so this is a parsing/offset bug, not a missing feature.

Root cause

read_element pads every element to an 8-byte boundary, but scipy and MATLAB do not pad miCOMPRESSED elements. They write each variable as its own miCOMPRESSED element, so the padded advance over-shoots into the next element's zlib stream and mis-reads its tag.

Instrumented walk on a 4-variable scipy file:

off=0/248   type=15 (miCOMPRESSED) size=43   <- variable 1 (decompresses + reads fine)
off=56/248  type=0  size=30720  <- ERROR

Variable 1 is 8 (tag) + 43 (data) = 51 bytes; the code jumps to 56, but the next variable's zlib tag is actually at byte 51 (bytes 51..56 are the next tag + start of its stream, not padding).

Fix

For MI_COMPRESSED, advance to data_end and skip any trailing zero padding (top-level element tags are never 0x00, so this also tolerates writers that do pad). All other element types keep the 8-byte padding.

Verification

• Real StaMPS ps2.mat (75 MB, MATLAB v5 compressed): reads, bperp -> (279, 1).
• StaMPS parms.mat (v5 struct): reads.
• 4-variable scipy savemat(..., do_compression=True): reads all four.
• cargo test -p pystamps-mat: 12/12 green, including a new regression test reads_multiple_unpadded_compressed_elements. The existing reads_compressed_mat_v5_elements modeled a single padded blob, so it never exercised the multi-element walk that real per-variable compression produces.

Note: native_stage6.rs::mat_v5_variable_shape returns Ok(None) on compressed top-level elements, so the compressed read path goes through MatData::read (fixed here); its mat_element_header carries the same padding assumption and would need the same fix if it ever walks compressed elements.

🤖 Generated with Claude Code

https://claude.ai/code/session_019PjQ6amGpaSTEbAv368SUc