Store SSMC text without char-matrix padding

[BenBrock]

• notes is now a single \n-joined string instead of a JSON array of blank-padded rows; aux text datasets are stripped row by row, as sstextwrite already does for MM and RB.
• Trailing blanks are stripped except on the widest row when that row is what carries the char matrix width, since readers rebuild with char(). The round trip stays exact, so ssexport's isequal check keeps passing and nothing in SuiteSparse needs to change.
• binsparse_write_ssmc_problem.m: new strip_text_rows (replaces char_rows), used by both text_block and write_string_dataset.
• binsparse_to_ssmc_problem.m: text_rows splits embedded line terminators and never deblanks; old files with padded row arrays still read identically.
• Of the 1863 matrices with notes, 1825 come out with no trailing blanks at all; the other 38 keep padding on one line (Mycielski/mycielskian*, Schenk/nlpkkt*, Cylshell/*, GHS_psdef/s3dk*, LPnetlib/lp_blend, lp_finnis, lp_perold, McRae/ecology1, Luong/photogrammetry).
• All binding tests pass, plus new assertions for both cases; ssexport(ids, 'check', '', {'MM','RB','BSP'}) passes on [2758 603 7 1474 23].

Not fixed here: MM and RB lose a column from aux text whose widest row is padded (sstextwrite deblanks, sstextread rebuilds narrower). Latent — no collection matrix has such aux today — and the fix belongs in SuiteSparse.