docs: record what the clip filter removes on the mitochondrial side

The MT-origin reads dropped by the 300-bp filter on the human runs are long reads
with noise tails (mitochondrial concatemers, unalignable sequence, or 93 to 341 bp
low-confidence hits at recurrent low-complexity coordinates), not NUMT reads.
Verified on the whole-genome records with mt_chimera_check.sh and mt_tail_detail.sh,
which are added here.

Co-Authored-By: Claude Fable 5.1 <[email protected]>