Move package usage docs to docs/package and slim .github/README.md to quickstart

[Copilot]

The current package usage documentation was concentrated in .github/README.md, making it hard to maintain and navigate by module. This PR relocates and expands usage docs under docs/package/, then simplifies the README to a quickstart + references.

• Package docs split by module

  • Added dedicated usage guides under:
    • docs/package/layers/{index,shaping,embedding,transformer}.md
    • docs/package/metrics/{index,group_accuracy}.md
  • Migrated and refreshed examples for:
    • shaping (Divide, Merge)
    • embedding (TokunEmbedding, RotaryPositionalEmbedding)
    • transformer (FeedForwardNetwork, CachedMultiHeadAttention)
    • grouped metrics (CategoricalGroupAccuracy, BinaryGroupAccuracy, RawGroupAccuracy)
  • Aligned examples/arguments with current src/mlable/* APIs and expected usage patterns shown in tests.

• README cleanup

  • Replaced long-form API sections in .github/README.md with:
    • install command
    • short quickstart snippet
    • links to the new module docs in docs/package/.

• Docs navigation updates

  • Updated docs/package/index.md with direct links to newly added guides.
  • Updated docs/index.md to reference docs/package/ structure.
  • Updated docs/todo.md to reflect that metrics API examples now exist in package docs.

import tensorflow as tf
import mlable.layers.shaping
import mlable.metrics

x = tf.ones(shape=(2, 4, 8))
y = mlable.layers.shaping.Divide(axis=-1, factor=4, insert=True)(x)

metric = mlable.metrics.CategoricalGroupAccuracy(group=4)
score = metric(
    y_true=tf.one_hot([[1, 2, 3, 4]], depth=8),
    y_pred=tf.one_hot([[1, 2, 0, 4]], depth=8),
)

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.