A bug in mesa for older AMD Silicon Island (SI) chipsets causes Firefox
(Nightly) to expose a latent bug that causes an immediate abort() crash.
Upstream 25.0.7 was supposed to fix it but it appears the fix is incomplete.

https://gitlab.freedesktop.org/mesa/mesa/-/work_items/13072

https://bugzilla.mozilla.org/show_bug.cgi?id=2083972

Mozilla analysis suggests further cherry-picks might fully resolve this
based on an OpenBSD bug caused by the same issue:

https://www.mail-archive.com/misc@openbsd.org/msg194540.html

I am about to test these on top of 25.0.7-2+deb13u1.

I've now confirmed a fix with 1 revert and 2 cherry-picks from upstream:

I started from Salsa's debian/25.0.7-2, added 25.0.7-2+deb13u1 NMU
changes on top, and then used:

git revert 2d525d5e3401555d97fe827f8db5bfed8887dc33       # radeonsi: always lower alu bit sizes
git cherry-pick f034aa9cd31ea4aa367f8ed58bd270026abf748e  # radv: don't use bit_sizes_int to skip nir_lower_bit_size
git cherry-pick 33b5d8b2ec26904ec1c331dd64ee04e351fd8e0e  # radeonsi: always lower alu bit sizes

After confirming these fix the issue I converted them to debian/patches
and am including a debdiff here.