Add dedicated KeyValueCache documentation page with worked examples

[Copilot]

This PR adds first-class documentation for KeyValueCache, comparable to the existing Scorer/Indexer/Retriever cache docs. It introduces a dedicated page with practical usage patterns and wires it into the main docs navigation.

• New KeyValueCache docs page

  • Added pyterrier_caching/pt_docs/key-value-cache.rst.
  • Covers:
    • what KeyValueCache is for (general transformer output caching by key columns),
    • caveats around key design and cache validity,
    • behavior when backing transformer is absent on cache miss.
  • Includes API sections for:
    • pyterrier_caching.KeyValueCache
    • pyterrier_caching.Sqlite3KeyValueCache

• Worked examples

  • Added example for caching an LTR-style features output keyed by ['query', 'docno'], with features stored as NumPy array values.
  • Added example for caching multiple value columns (feature-style outputs) for the same key, including a query-derived question-word feature.

from pyterrier_caching import KeyValueCache

cached_features = KeyValueCache(
    "path/to/feature-cache",
    feature_builder,
    key=["query", "docno"],
    value=["qlen", "has_question_word"],
)

• Docs discoverability updates
  • Updated pyterrier_caching/pt_docs/index.rst to include key-value-cache in the toctree and component list.
  • Updated README.md overview links to include the new KeyValueCache documentation page.

[cmacdonald]

To better match the PyTerrier LTR data model, column should nane be features, not feature.
also the value should be an numpy array, not a single value.

[Copilot]

Implemented in commit 9f174e5: the example now uses features (not feature) and populates it with NumPy arrays (np.array([...], dtype=np.float32)), with value=['features']. Screenshot: updated docs snippet is visible in the PR file diff for key-value-cache.rst.