Minor changes to make pandas 1.4.0 work

[zaneselvans]

A PR to fix anything that Pandas 1.4.0 breaks. Draft for now because I am running tox -e nuke locally.

The full ETL + data validation works with this PR, but it's also giving a warning about converting between naive & timezone aware datetime objects using df.astype() and I'm concerned that using apply_pudl_dtypes() across the board may be messing up timezone info. But I think that's a separate issue (See: #1423)

[codecov]

Codecov Report

Merging #1421 (94dd108) into dev (ab419c9) will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##              dev    #1421      +/-   ##
==========================================
+ Coverage   83.46%   83.48%   +0.02%     
==========================================
  Files          64       64              
  Lines        6930     6937       +7     
==========================================
+ Hits         5784     5791       +7     
  Misses       1146     1146              

Impacted Files	Coverage Δ
src/pudl/analysis/mcoe.py	92.73% <ø> (ø)
src/pudl/extract/epacems.py	97.67% <ø> (ø)
src/pudl/output/eia860.py	65.88% <ø> (ø)
src/pudl/extract/eia860m.py	100.00% <100.00%> (ø)
src/pudl/extract/excel.py	95.83% <100.00%> (+0.04%)	⬆️
src/pudl/glue/ferc1_eia.py	80.00% <100.00%> (+0.13%)	⬆️
src/pudl/output/ferc714.py	93.25% <100.00%> (ø)
src/pudl/transform/eia.py	95.51% <100.00%> (ø)
src/pudl/transform/eia861.py	97.01% <100.00%> (+0.03%)	⬆️
src/pudl/transform/eia923.py	93.97% <100.00%> (ø)
... and 1 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ab419c9...94dd108. Read the comment docs.

[bendnorman]

Looks good! It seems like the timezone thing isn't a huge issue? Maybe we address it in a separate PR.

1.4.0 added support for reading csvs with pyarrow. I'm going to experiment running CEMS using the pyarrow engine for pd.read_csv()

[zaneselvans]

Yeah, I created another issue for the timezone thing, but I do think we should investigate it and come up with a way to treat the timezone aware/naive columns appropriately.

We need to give pd.read_csv() a dtype template of some kind when it's reading in the CEMS files, otherwise when the column is empty it gets assigned one type, and when it's got data it's assigned another type, and when the dataframes are concatenated together you get inhomogeneous "object" types which slow down subsequent operations. But it's slightly more complicated because not all of the possible columns appear in every file, so the dtype dictionary has to somehow be sensitive to that. I fiddled with it a bit and gave up.