Add QDM output Zarr priming, region writing, metadata, multi-year processing

[brews]

This PR rewrites the QDM command prototypes, breaking backward compatibility.

Changes include:

• New command for priming a Zarr Store so it can be written to regionally by independent processes. For example:

dodola prime-qdm-output-zarrstore \
  --simulation "gs://path/to/file/we/want/to/biascorrect.zarr" \
  --variable "tasmax" \
  --years "2015,2100" \
  --out "gs://path/to/where/we/want/biascorrected/to/go.zarr" \
  --zarr-region-dims "lat"

We should then be set to write output to that zarr store, regionally, with regions delimited along the "lat" dimension.

• Write to regions of a primed Zarr Store aggressively with dodola apply-qdm --out-zarr-region. For example, --out-zarr-region "lat=0,2" will write QDM-adjusted output to slice(0, 2) along the "lat" dimension. Note this is written aggressively (with mode="a"), so use caution not to corrupt existing data. Use this with the --selslice and --iselslice arguments to filter input data.

• dodola apply-qdm now expects to be applied to multiple years, rather than single years. It does this through a --years argument. It's used like --years=2015,2100. This should help work run faster and more efficiently on spatially regional subsets consuming the entire input time series.

• Root and variable attributes are now copied from input data to output QDM-adjusted datasets. QDM output should now also match the dtype of the input dataset variable.

• QDM's sim_q are now output as a variable instead of a coordinate. These are now output by default.

• apply-qdm and prime-qdm-output-zarrstore both accept one or more --new-attrs options. This is a fast way of adding simple metadata to output. Each of these take key=value pairs that are merged into the output Dataset's root attrs before being written.

• Logging is slightly more chatty on INFO, especially with respect to data slicing.