mesa: upgrade 25.3.5 -> 26.0.0

mesa: upgrade 25.3.5 -> 26.0.0

See https://docs.mesa3d.org/relnotes/26.0.0.html for the changelog and
sha256sum.

This removes:
- 0001-ethosu-drop-file-names-from-the-generated-file.patch, as it is
  merged in commit cdb6468c53ef ("ethosu: drop file names from the
  generated file"),
- 0001-gfxstream-don-t-dump-genvk.py-args-to-generated-file.patch, as it
  is merged in commit 4315c28739dc ("gfxstream: don't dump genvk.py args
  to generated files"),
- 0002-rocket-drop-file-names-from-the-generated-file.patch, as it is
  merged in commit 7a3bfd1f7913 ("rocket: drop file names from the
  generated file"),

0001-freedreno-don-t-encode-build-path-into-binaries.patch is updated to
latest version in the merge request. Unfortunately, the do_patch task
with the patch file directly downloaded from the merge request fails to
apply it due to patch-fuzz. Instead, I downloaded the file locally and
applied it with git-am and use git-format-patch to generate one that
do_patch doesn't complain about.

Since commit b72d347d1c3b ("meson: crocus and intel_hasvk now require
clc"), both crocus and intel_hasvk require clc, so let's make this
dependency explicit by requiring libclc PACKAGECONFIG to enable those
drivers.

KosmicKrisp is a new driver but my understanding is it's for MacOS only
and we're not generated that OS, so skip adding a PACKAGECONFIG for it.

r600 now doesn't require LLVM anymore since commit ebf14544102e ("r600:
remove LLVM dependency") but for some reason we never added the
dependency so no need to remove something that isn't there.

There's a new option vulkan-manifest-per-architecture (default true)
which controls whether the target machine architecture must be part of
the Vulkan ICD filenames. The meson feature description states:

If false, all architectures share a single JSON manifest for example
lvp_icd.json, referencing the library by its basename. (Recommended for
Unix OS distros installing into /usr.)

Considering we install the Vulkan ICDs into /usr/share, let's make this
false by default. Also, we're only building for one architecture so this
probably doesn't change much except renaming.

Signed-off-by: Quentin Schulz <quentin.schulz@cherry.de>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>