smaller images: remove docs files and build artifacts, consolidate layers

[jameslamb]

Contributes to rapidsai/shared-workflows#505

A couple of small changes aimed at reducing image sizes (and therefore pull times in CI):

• consolidates some RUN steps (fewer layers = smaller image)
• removes man pages and other docs
  • adds --nodocs to every dnf install
  • manually removes /usr/share/man and similar
• consolidates some package-install calls (fewer solves = faster builds)
• changes specific to rockylinux8 ci-wheel images:
  • removes 450+ MB (uncompressed) of left-behind OpenSSL build artifacts in /tmp
  • removes gcc install (we're already installing gcc-toolset-* packages)

Notes for reviewers

Impact of these changes

For CUDA 12.9.1, Python 3.11, x86_64, rockylinux8:

image	before	after
ci-conda	750.78 MB (link)	748.72 MB (link)
ci-wheel	5.24 GB (link)	5.14 GB (link)

So not a huge change, but 100MB compressed cut out of ci-wheel is decent! Let's merge this and see if pull times improve.