v0.0.2: JIMM has sharding!

With this release, jimm now has sharding working correctly, with gradient checkpointing allowing for the model to scale effectively.

What's Changed

• feat: fix the sharding to auto compile the model to devices in a sharded manner, rather than making the user do so. Unsure if this is good design or not by @pythoncrazy in #23
• Add sharding examples for the clip models with both nnx.jit and nnx.shard_map by @pythoncrazy in #24
• Sharding scaling by @pythoncrazy in #27
• fix: fixed the memory requirements by using local arrays then converting them to global arrays etc by @pythoncrazy in #28
• fix: add the tpu-info dependency to view the system info more easily by @pythoncrazy in #29
• feat: use local images for the tests, and having sharding for the tests by @pythoncrazy in #33
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #32

Full Changelog: 0.0.1...0.0.2