Caching problem with _inv_cdf_func

[sgherbst]

It is currently not possible to have different settings for num_sigma, lfsr_width, func_order, and func_numel for Gaussian noise generators due to how _inv_cdf_func is cached in MixedSignalModel:

msdsl/msdsl/model.py

Lines 373 to 381 in 58964b3

	if self._inv_cdf_func is None:
	inv_cdf = lambda x: truncnorm.ppf(x, -num_sigma, +num_sigma)
	domain = [apply_compression(0), apply_compression((1<<(lfsr_width-1))-1)]
	self._inv_cdf_func = self.make_function(
	lambda x: inv_cdf(invert_compression(x)/(1<<lfsr_width)),
	domain=domain,
	order=func_order,
	numel=func_numel
	)

We should make the caching more flexible so that it knows about the values of those four settings for each cache entry.