Use libmamba solver

[austin3dickey]

Fixes #103. Tells conda to use the mamba solver. I thought this was easier than going through and installing mamba on all our machines.

[austin3dickey]

I kicked off a test run on every machine to ensure they all work. Might take a while.

[austin3dickey]

So far tests are running well; I think I'll need to manually update conda on ursa-thinkcentre-m75q. But for the two that have finished so far:

• ec2-t3-xlarge-us-east-2: avg 80m down to 33m
• test-mac-arm: avg 115m down to 107m

[austin3dickey]

• arm64-t4g-linux-compute: 160m -> 160m
• ec2-m5-4xlarge-us-east-2: 160m -> 160m
• ursa-i9-9960x: 160m -> 147m
• arm64-m6g-linux-compute: 255m -> 257m

So we don't see a speedup across the board; mostly just for the ec2-t3-xlarge-us-east-2 machine. But I think this should maybe help with https://github.com/voltrondata-labs/arrow-benchmarks-ci/issues/174.

[austin3dickey]

Still seeing symptoms of #174 after I upgraded conda on the machine and tested this PR. Weird. For now I added a retry to that step in particular.