Consider relaxing default tolerance for truncate_small_values

[billsacks]

The default tolerance for truncate_small_values is 1e-13. This is proving to be too stringent in some cases - e.g., see #2457 . Currently there is a mechanism to override the default tolerance on a case-by-case basis, but it might be better to just relax the tolerance to make us less likely to run into issues in the future. I'd suggest at least relaxing it to 1e-12, and possibly 1e-11 or 1e-10. The danger of relaxing it too much is that we could truncate values to 0 that are really meant to be non-zero. But even with 1e-10 this means that a state variable was just reduced by 10 orders of magnitude, so setting it to 0 in that case seems unlikely to do too much harm scientifically, unless this is happening a lot in which case I guess it's possible that we could get accumulating conservation issues.

If we make this change, we should remove settings of custom_rel_epsilon that are currently using values smaller than the new default. e.g., we should remove settings of custom_rel_epsilon=1e-12 if the new default is 1e-12, 1e-11, 1e-10 or bigger.

[billsacks]

I realized a possible issue with relaxing the tolerance too much – or with how this truncation is done in general – and this led me to think of a different, probably more robust way to handle this truncation - see #2459 .

I'm keeping that as a separate issue because in the shorter-term I think it could make sense to just relax the tolerance a bit, but longer term I think that more substantial rework could be warranted.

[ekluzek]

This was completed as part of #2457 so closing.