add retry framework to burn

[zingale]

This adds the ability to retry the burn in a zone if it fails, without having
to return to the application code.

It provides a few new parameters:

• use_burn_retry
• retry_swap_jacobians (this will use the other Jacobian type for the retried burn)
• retry_rtol_spec, retry_rtol_enuc (the relative tolerance to use on the retry)
• retry_atol_spec, retry_atol_enuc (the absolute tolerance to use on the retry)

This works well near NSE where sometimes the other Jacobian type can get past
a stubborn burn.

The cost of this change is that we need to store an old copy of the burn_t, but
this is templated out if we don't run with use_burn_retry=1

[zingale]

this seems to work, although it comes with the overhead of a backup burn_t

[zingale]

CPU tests pass: http://groot.astro.sunysb.edu/Microphysics/test-suite/gfortran/2022-09-07-008/index.html

[zhichen3]

if both of them are false, then wouldn't it still swap the jacobian?

[zingale]

do you mean is !(false && false) == true?

[zingale]

indeed, I think we need to rewrite this logic

[zingale]

wait... I think it is okay.

This will evaluate to true is either or both of is_retry and retry_swap_jacobian are false, in which case it does not swap the jacobian.

only if both is_retry = true and retry_swap_jacobian = true does it do:

be.jacobian_type = (jacobian == 1) ? 2 : 1;

which is the swap.

right?

[yut23]

I did a quick truth table in a spreadsheet, and the logic is correct right now. It's definitely not obvious at first glance, though. I think swapping the if/else branches would help.

[zingale]

agreed. It wasn't obvious to me either, and I wrote it ;)

[zingale]

updated