Minor: use venv in benchmark compare

[tmi]

Which issue does this PR close?

Closes #10022

(Note: this PR follows up on #10070 which worked on the same issue but looks like got kinda stale)

Rationale for this change

Make it more convenient to run bench.sh compare

What changes are included in this PR?

Updates to bench.sh to include a new command venv, and edit the compare command to use the respective venv

Separates requirements.txt in case compare.py would ever need something else

Are these changes tested?

I tested by hand ./bench.sh compare and ./bench.sh venv, in multiple scenarios:

• create empty venv with bench.sh, then run compare
• have existing venv, update with bench.sh venv, then run compare
• have no venv but install rich into system, and just run compare
  In all cases compare works correctly.

Are there any user-facing changes?

No -- if the compare worked for the user before, it will still do, regardless of whether they used venv or not.
If it didn't work, then the user would see improved error message, run ./bench.sh venv without a need to configure any env var, and the new venv would be created in a gitignored folder in the repo

[alamb]

Looks good to me -- thank you @tmi

[alamb]

I merged up to get a clean CI run

[comphead]

lgtm thanks @tmi for the nice first contribution