GH-41964: [CI][C++] Clear cache for mamba on AppVeyor

[kou]

Rationale for this change

It seems that mamba may use invalid download URL when there are invalid caches.

What changes are included in this PR?

Clear caches explicitly.

Are these changes tested?

Yes.

Are there any user-facing changes?

No.

• GitHub Issue: [CI][C++] Fail to install libboost-headers with mamba on AppVeyor #41964

[github-actions]

⚠️ GitHub issue #41964 has been automatically assigned in GitHub to PR creator.

[pitrou]

Wow. Is this something that should be reported to the mamba issue tracker?

[pitrou]

+1, thanks a lot @kou

[kou]

Is this something that should be reported to the mamba issue tracker?

Hmm... Maybe no...?
It seems that this is related to our AppVeyor environment.

This is the last succeeded job before this problem is happen. Note that mamba 1.5.8 was selected:

https://ci.appveyor.com/project/ApacheSoftwareFoundation/arrow/builds/49866639?fullLog=true#L90

C:\projects\arrow>conda install -q -y -c conda-forge mamba python=3.10   || exit /B 
Collecting package metadata (current_repodata.json): ...working... done
Solving environment: ...working... done
## Package Plan ##
  environment location: C:\Miniconda38-x64
  added / updated specs:
    - mamba
    - python=3.10
The following packages will be downloaded:
    package                    |            build
    ---------------------------|-----------------
...
    libmamba-1.5.8             |       h3f09ed1_0         3.4 MB  conda-forge
    libmambapy-1.5.8           |  py310h04f2035_0         330 KB  conda-forge
    mamba-1.5.8                |  py310hd9d798f_0          67 KB  conda-forge

This is the first failed job by this problem. Note that there are some solving failed messages and mamba 0.24.0 was selected. (mamba is updated to 1.5.8 later):

https://ci.appveyor.com/project/ApacheSoftwareFoundation/arrow/builds/49867058#L88

C:\projects\arrow>conda install -q -y -c conda-forge mamba python=3.10   || exit /B 
Collecting package metadata (current_repodata.json): ...working... done
Solving environment: ...working... failed with initial frozen solve. Retrying with flexible solve.
Solving environment: ...working... failed with repodata from current_repodata.json, will retry with next repodata source.
Collecting package metadata (repodata.json): ...working... done
Solving environment: ...working... done
## Package Plan ##
  environment location: C:\Miniconda38-x64
  added / updated specs:
    - mamba
    - python=3.10
The following packages will be downloaded:
    package                    |            build
    ---------------------------|-----------------
...
    libmamba-0.24.0            |       h81a967f_1         2.9 MB  conda-forge
    libmambapy-0.24.0          |  py310hd80b381_1         527 KB  conda-forge
...
    mamba-0.24.0               |  py310h9376f3e_1          64 KB  conda-forge

I'm not sure whether this is related to this problem or not but our AppVeyor environment may be related. Because cleaning cache solved this problem.

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 4 benchmarking runs that have been run so far on merge-commit 67c6df1.

There were no benchmark performance regressions. 🎉

The full Conbench report has more details.