POC for zarr support - DO NOT MERGE

[JSKenyon]

• [ ] Tests added / passed
• [ ] Fully documented, including HISTORY.rst for all changes and one of the docs/*-api.rst files for new API

Howto run test cases and lint the code base

```bash $ py.test --flake8 -v -s tricolour ``` If you encounter flake8 failures, a quick way to correct this is to run `autopep8` and `flake8` again. ```bash $ pip install -U autopep8 $ autopep8 -r -i tricolour $ flake8 tricolour ```

Howto build the documentation

To build the docs locally: ```bash $ pip install -r requirements.readthedocs.txt $ cd docs $ READTHEDOCS=True make html ```

This PR replaces xds_from_ms, xds_from_table and xds_to_table with their xds_to/from_storage_* equivalents. As part of this change I had to unpin basically all the dependencies i.e. this is not safe to merge. The reason for this change is to establish whether we can use Tricolour on AWS using zarr backed data in s3.

This runs through on a toy example on my laptop but there are still some sharp edges. The most important of these is that once a dataset has been converted to zarr, we no longer have the ability to reorder it via TAQL i.e. for now we will assume that any conversion will result in time-ordered data. This could be improved but shouldn't be necessary for POC experiments.

[ratt-priv-ci]

Can one of the admins verify this patch?