- Add robust csv parsing and allow creating an expression matrix from it (#3)
- Always treat matrix row/col names as str
- Validate matrix name, raise UserError if matrix has duplicate
  index/col or if it has an invalid float value

- Add baits file parsing (#3)

- Remove yaml !include syntax (#3)
- Wrap yaml errors in UserError

Refactor:

- Eliminate libyaml dependency
- Remove unused dependencies
- Remove test coverage tools
- Install pylint dev dependency
- Update to pytil 8
- Use logging directly instead of using a logger
- Update my email address

- Bump version to 3.0.0.dev
- Update dependencies and only constrain them to a minimum version, no
  pinning, not even to the major version

Previous code did not ignore it.

This is normal for some inputs, e.g. a constant input (1, 1, 1).

- Whole API ref on one page. Use autosummary instead of listing objects
  in module manually.
- Use intersphinx
- Use PEP484 instead of our own type language
- Rephrase and consistently end with '.'

E.g. 4.1.0 lacks [test] extra, which breaks installation

Narrowing scope of library to simple parsing. Our algorithms
usually can keep everything in memory and thus don't need a database

instead, use Nextflow, Snakemake or a similar tool

This reverts commit 23355fb5.

Use custom pool with enough threads to run and await `max_cores_used`
amount of jobs.

Solution: Use separate thread pool for cancelling, such that threads are
always available to cancel with.