Fix Gibbs sampler bug and add initial_amplitude parameter, and fix misc. MPIArray warnings

[sjforeman]

This PR has turned into a Frankenstein of various edits, but they are all relatively minor:

• Fix a few MPIArray warnings I've found in various tasks.

• Fix the following bug in DelaySpectrumEstimatorBase: if the input container has a large number of "baselines" with zero data or weights, such that there is an MPI rank that only contains these baselines, the routine that initializes the random number generator will never be called on this rank, because self.rng is only called in the argument of delay_spectrum_gibbs() in the main loop:

  draco/draco/analysis/delay.py

  Lines 671 to 716 in 1603553

  	for lbi, bi in delay_spec.spectrum[:].enumerate(axis=0):
  	self.log.debug(f"Delay transforming baseline {bi}/{nbase}")
  	# Get the local selections
  	data = data_view[lbi].view(np.ndarray)
  	weight = weight_view[lbi].view(np.ndarray)
  	# Mask out data with completely zero'd weights and generate time
  	# averaged weights
  	weight_cut = (
  	1e-4 * weight.mean()
  	) # Use approx threshold to ignore small weights
  	data = data * (weight > weight_cut)
  	weight = np.mean(weight, axis=0)
  	if (data == 0.0).all():
  	continue
  	# If there are no non-zero weighted entries skip
  	non_zero = weight > 0
  	if not non_zero.any():
  	continue
  	# Remove any frequency channel which is entirely zero, this is just to
  	# reduce the computational cost, it should make no difference to the result
  	data = data[:, non_zero]
  	weight = weight[non_zero]
  	non_zero_channel = channel_ind[non_zero]
  	spec = delay_spectrum_gibbs(
  	data,
  	ndelay,
  	weight,
  	initial_S,
  	window=self.window if self.apply_window else None,
  	fsel=non_zero_channel,
  	niter=self.nsamp,
  	rng=self.rng,
  	complex_timedomain=self.complex_timedomain,
  	)
  	# Take an average over the last half of the delay spectrum samples
  	# (presuming that removes the burn-in)
  	spec_av = np.median(spec[-(self.nsamp // 2) :], axis=0)
  	delay_spec.spectrum[bi] = np.fft.fftshift(spec_av)

  , and this will always be skipped. The solution is to explicitly initialize the RNG on all ranks before entering the loop, by calling rng = self.rng.

• Allow the user to specify the amplitude of the initial power spectrum guess in DelaySpectrumEstimator and DelaySpectrumEstimatorBase.