Remove hard-coded model parameter indexing in MCMC calibration

[btobers]

Currently the mcmc module is hard-coded to expect the model params in the order of [tbias, kp, ddfsnow] - based on the order of the priors and the initial guesses. It'd be better to utilize matched-key dictionaries.

Originally posted by @btobers in #176 (comment)

[btobers]

Some places where the order of the model parameters are hard-coded:

PyGEM/pygem/bin/run/run_calibration.py

Lines 2181 to 2186 in a0629ad

	initial_guesses = torch.tensor(
	(
	tbias_mu,
	kp_gamma_alpha / kp_gamma_beta,
	pygem_prms['calib']['MCMC_params']['ddfsnow_mu'],
	)

PyGEM/pygem/mcmc.py

Lines 260 to 263 in a0629ad

	kwargs = {
	'tbias': m[0],
	'kp': m[1],
	'ddfsnow': m[2],