Add pathogen-embed to the base image

[huddlej]

Description of proposed changes

Adds pathogen-embed to base image, so we can use its tools for analysis of reassortment in our influenza builds.

Related issue(s)

nextstrain/conda-base#71

Checklist

• Checks pass

[victorlin]

(non-blocking)

Noting from the run summary that this added 8.5 minutes to the build time:

[linux/arm64] RUN pip3 install pathogen-embed==2.0.0     | 496.7s (61.9%)  ████████████████████
[linux/amd64] RUN pip3 install pathogen-embed==2.0.0     | 17.0s (2.1%)    ▋

I'm not worried since this dependency is pinned, meaning the cached result will be used most of the time and that the increased build time would only be noticeable during (infrequent) cache misses.

[huddlej]

This one change also adds ~57 MB to the image file size... 😕

I expected an increase, since we need to install scikit-learn, HDBSCAN, and UMAP, but it is kind of a bummer.

[huddlej]

Even though the CI passed, this PR doesn't currently work as expected. When I download the image for this branch and run a Nextstrain shell like so, the image is missing the executables for pathogen-distance, pathogen-embed, etc.:

$ nextstrain shell --docker --image docker.io/nextstrain/base:branch-add-pathogen-embed .
~/build $ which pathogen-distance
~/build $ which pathogen-embed
~/build $ 

However, the Python package is installed, since I can load the Python REPL from the Nextstrain shell and import it:

~/build $ python
>>> import pathogen_embed
>>>

[huddlej]

I forgot this step:

docker-base/Dockerfile

Lines 433 to 452 in 3f15ce3

	# Add installed Python scripts that we need.
	#
	# XXX TODO: This isn't great. It's prone to needing manual updates because it
	# doesn't pull in scripts which got installed but that we don't list. Consider
	# alternatives (like installing the deps into an empty prefix tree and then
	# copying the whole prefix tree, or using pip's installed-files.txt manifests
	# as the set of things to copy) in the future if the maintenance burden becomes
	# troublesome or excessive.
	# -trs, 15 June 2018
	COPY --from=builder-target-platform \
	/usr/local/bin/augur \
	/usr/local/bin/aws \
	/usr/local/bin/envdir \
	/usr/local/bin/nextstrain \
	/usr/local/bin/pangolin \
	/usr/local/bin/pangolearn.smk \
	/usr/local/bin/scorpio \
	/usr/local/bin/snakemake \
	/usr/local/bin/treetime \
	/usr/local/bin/

Fixing in the next commit.

[huddlej]

I confirmed that copying the scripts in the last commit properly provided each of the pathogen-embed tools. I additionally tested that these tools worked in this image with pathogen-embed's cram tests (from the pathogen-embed repo) like so:

$ nextstrain shell --docker --image docker.io/nextstrain/base:branch-add-pathogen-embed .
~/build $ python -m pip install cram
~/build $ ~/.local/bin/cram --shell=/bin/bash tests
.............................
# Ran 29 tests, 0 skipped, 0 failed.
~/build $

I'll plan to merge this PR and the sibling conda-base PR on Monday.