Pfhedge

* ENH: Replace LazyLinear in MLP with Linear (close 83) (84)

* DOC: Elaborate examples

* DOC: Move `Hedger` to `nn` (close 71) (73)

* DOC: Make "E" for an expectation value boldfaced (close 69) (74)

* DOC: Fix minimal example (close 68) (75)

* REF: Use `register_buffer` in `save_prev_output` (close 70) (76)

* Update to PyTorch 1.9.0

* ENH: REF: Make default `dt=1/250`, `maturity=20/250` (close 58) (61)

* REF: `Move nn.bs` to `nn.modules.bs` (close 55)

* REF: Move `Hedger` to `nn.modules` (close 56) (62)

* REF: Rename `features` to `inputs` (close 54) (63)

* ENH: Put batch to the first dim, time steps to the last dim (close 57) (64)

* REF: Make use of `torch.testing.assert_close` (close 60) (65)

* REF: Let `MultiLayerPerceptron` inherit `Sequential` (close 41)

* DOC: Modify docstring to Google Style (50)

* DOC: Create Sphinx Documentation (close 39) (42)

* ENH: Add `dtype` and `device` to `__repr__` of instruments (close 28) (35)

* ENH: Add `dtype` and `device` to parameters of instruments (37)