v0.1.0: First release

This release includes:

• Basic module system: pax.Module.
• Basic leaf node: Only two supported nodes: pax.State and pax.Parameter which are subtypes of pax.tree.Leaf.
• Basic random state: a global rng_key is stored at pax.rng.state._rng_key. Call pax.next_rng_key() to generate a new rng_key.
• Basic optimizer: pax.Optimizer.
• nn modules: nn.Linear, nn.BatchNorm, nn.LayerNorm. nn.Conv1d, nn.Conv2d.
• Convert dm-haiku module to pax.Module with pax.haiku.from_haiku.
• Convert optax optimizer to pax.Optimizer with pax.optim.from_optax.
• An example training RNN language model on TPU.
• An example training a MNIST classifier with checkpoint.