Merge pull request #56 from ehfd/pr

> An older TensorFlow version then 2.18.0 is blocked, obviously, and key capabilities in JAX breaks with TensorFlow Probability 0.25.0 at around version 0.4.24 or 0.4.23 (I have tested so). So decreasing the lower limits is risky, and should be kept, whether or not pip enforces it. But the upper limits - I doubt there will be issues.

Very well, I may misunderstood, but we agree.

> I think if there is an issue with the upper bounds somehow, we can limit it when there is an issue, since pip would under the same problem.

It's not that easy, and the argument also works the other way around: you can relax it once we know it works. I am saying that it's not that easy, because how do you know it broke because of a mismatched version? A user may randomly bumps into it. But that's not a version error. That's a random error. And the user tracks it down, may first thinks it's him. If he is a power user, he can track it down and, fair enough, report it _maybe_ in TFP (or here). But there are a lot of ifs. Most notably, if will cause _hard to understand_ bugs for users.


But we agree, LGTM!

.ci_support/linux_64_.yaml

@@ -1,19 +1,12 @@
-c_compiler:
-- gcc
-c_compiler_version:
-- '12'
 cdt_name:
-- cos6
+- conda
 channel_sources:
 - conda-forge
 channel_targets:
 - conda-forge main
-cxx_compiler:
-- gxx
-cxx_compiler_version:
-- '12'
 docker_image:
-- quay.io/condaforge/linux-anvil-cos7-x86_64
-zip_keys:
-- - c_compiler_version
-  - cxx_compiler_version
+- quay.io/condaforge/linux-anvil-x86_64:alma9
+python_min:
+- '3.9'
+tensorflow:
+- '2.16'