pin to specific version in test and docs builds

[jameslamb]

Contributes to #115

To test the release workflow, I did this kind of weird thing where. I pushed a tag v24.08.00 when there was already an "earlier" tag v24.09.00.dev in the repo.

That all worked fine... but pulled in the wrong version in the docs build. Because legate-boost is unpinned here:

legate-boost/ci/build_docs.sh

Lines 30 to 37 in 4151942

	rapids-mamba-retry install \
	--name docs-env \
	--override-channels \
	--channel "${RAPIDS_LOCAL_CONDA_CHANNEL}" \
	--channel legate \
	--channel legate/label/experimental \
	--channel conda-forge \
	"legate-boost=*=*_cpu*"

conda chose one with the highest version. That was 24.09.00.dev3, even for a workflow run that had produced 24.08.00. Oops.

This fixes that.

[seberg]

CI is choking on the version:

Looking for: ['legate-boost==24.09.00[build=*_cpu*]']
<snip>
The following package could not be installed
└─ legate-boost 24.09.00 *_cpu* does not exist (perhaps a typo or a missing channel).

But I don't really see why this is different from the tests. I guess we don't need the =*_cpu*, but does it make it not find dev versions?

[jameslamb]

I guess we don't need the =_cpu, but does it make it not find dev versions?

I think you're right! That == makes me thing that maybe conda is looking for exactly 24.09.00 (no dev versions or other modifiers).

I think that removing it might allow it to float.

I've also put us into a weird position for testing here by pushing tags to the repo like v24.08.00 that result in new packages having lower versions than older packages (v24.09.00.dev{n}), which is not something that's likely to happen during normal development. I was just trying to be careful in case we weren't able to delete packages from the legate channel, but I'm afraid I might have just made this more complicated than it needed to be 😬

[jameslamb]

I just pushed the following changes:

• remove *_cpu* tag in this script
• run rapids-generate-version first, to update the VERSION file from git tags

Those two together should hopefully fix this.

And once we get this all working, I can put up similar changes in legate-dataframe.

[jameslamb]

I just deleted those v24.08.00 / v24.08.01 tags and re-triggered CI here. I think the changes in this PR will be enough to get this working on pushes to main (which I'll test after we merge this).