ENH: accept tf.data.Dataset inputs

[adriangb]

Partially addresses #160 by enabling direct tf.data.Dataset inputs.

This mainly enables the use case where a user has an existing Dataset and Model and would like to use GridSearch or something on it, but most of the rest of the SciKeras and Sklearn features will be unavailable including:

• Encoding/decoding of labels to allow strings, obj, etc.
• Ability to dynamically adapt Model inputs/outputs to data

And this requires using BaseWrapper directly, or alternatively subclassing KerasClassifier or KerasRegressor to disable the transformers.

[codecov-io]

Codecov Report

Merging #166 (79620cd) into master (f4a9fef) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #166   +/-   ##
=======================================
  Coverage   99.55%   99.55%           
=======================================
  Files           6        6           
  Lines         672      679    +7     
=======================================
+ Hits          669      676    +7     
  Misses          3        3           

Impacted Files	Coverage Δ
scikeras/wrappers.py	99.45% <100.00%> (+0.01%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f4a9fef...79620cd. Read the comment docs.

[github-actions]

📝 Docs preview for commit 79620cd at: https://www.adriangb.com/scikeras/refs/pull/166/merge/