Fix for mcmc api transforms None for potential_fn

[ludkinm]

Fixes #2272.

This PR fixes a logic gap for assign self.transforms in the MCMC api.
It can be the case that both MCMC.transforms is None and MCMC.kernel.transforms is None
in which case MCMC.transforms should be set to an empty dict.
This bug appears when using HMC/NUTS with a potential_fn.

Note sure on the style guide for inline comments so will leave them as is.

[neerajprad]

LGTM, once the spacing issue is fixed. Thanks for fixing this!

[neerajprad]

I just pushed a fix to release a patch for 1.2 with this fix.

[neerajprad]

There are a few issues still with multiprocessing, I'll either fix it here or put out a separate PR.

[neerajprad]

ludkinm - Could you review the minor fixes that I made for #2272, and let me know if that's in line with your use case?

[fehiepsi]

LGTM, thanks @ludkinm and @neerajprad, especially for your comments about what is going on!

@neerajprad could you dismiss your previous review?

[neerajprad]

@fehiepsi - can we merge this? I would like to fix #2272 and release a patch update.

[fehiepsi]

@neerajprad your review is still in change request status, which does not allow me to merge.

Dismiss the review

[fehiepsi]

Sorry, I just know that I can dismiss your review :)

[ludkinm]

Sorry I have been away from the office so didn't respond.
@neerajprad thanks for your fix, it works in my use case too - my python is rusty and forgot that if statements are evaluated lazily.