more migrators skipping packages

[minrk]

this appears to be the same symptom as #5382 but with a different cause. Two different migrators seem to have skipped dependencies recenty:

1. mpich5 opened a PR for dolfinx_mpc before its dependency fenics-dolfinx, which is (rightly) waiting on its dependency adios2
2. petsc325 opened a PR for scifem which also depends on fenics-dolfinx, but has not yet received its PR.

Notably, in both cases fenics-dolfinx does not show up in the migrator, but it does show up in the graph as a dependent of both mpich and petsc. I'm not sure exactly how the bot decides if a dependent package correctly located in the graph should be excluded from migrators, but there seems to be something amiss there.

[minrk]

I discovered whily trying to debug the petsc migration that the petsc325 migrator has been closed out, with only one of its PRs having been merged. Are migrators supposed to close when all PRs have been opened, or when the migration has actually been completed?

[beckermr]

All PRs opened is when it closes.

[minrk]

Tracked it down: conda-smithy had deleted a variable from .ci_support files for Windows for some reason, leading to a failed parsing in the bot. So the fact that there was a bad render in parsing_error meant it got ignored from the migrator, rather than showing up with an error, despite the fact that it is correctly identified as a dependent package in the graph.

[minrk]

I believe this has been fixed for future migrators by a rerender. Should the error have been made visible somewhere, either in the migrator or on the feedstock?

[beckermr]

We do not surface parsing errors on nodes, so currently no. If a node cannot be parsed, then there is just not a lot we can do since we don't know what the feedstock does.