squash! Add trainable theta and discretization options

- Simplify A/B calculations
- More efficient cont2discrete implementation
- Rename train_theta to trainable_theta
- Store A/B as constants
- Remove A/B caching in zoh training
- Add test to make sure that zoh and euler produce (approximately)
  the same output
- Disable FFT if trainable_theta=True
- Update trainable_theta tests

CHANGES.rst

@@ -19,7 +19,7 @@ Release history
    - Removed
    - Fixed
 
-0.3.2 (unreleased)
+0.4.0 (unreleased)
 ==================
 
 **Added**
@@ -30,6 +30,12 @@ Release history
   are satisfied (no hidden-to-memory or memory-to-memory connections,
   and the sequence length is not ``None``). (`#40`_)
 
+**Changed**
+
+- The ``A`` and ``B`` matrices are now stored as constants instead of non-trainable
+  variables. This can improve the training/inference speed, but it means that saved
+  weights from previous versions will be incompatible. (`#40`_)
+
 .. _#40: https://github.com/nengo/keras-lmu/pull/40
 
 0.3.1 (November 16, 2020)