Booze-tools

Fix some nasties in the LR(1) pathways. Better tests now, too.
Improve the interface, too. Pretty output is prettier, and you can override to get a particular table method on the command line.

Adopted/allowed Py3.9. Also, experimenting with new 'arborist" module which probably hurts performance but begins to look like a platform for understanding a (semantic) parse tree.

Finished up during the Covid-19 scare of early 2020. Major new features include:

* Respectable error handling, with error-productions and an API for reporting error events -- at least on the deterministic side.
* The condensed parse-action table now includes ALL error entries, so bad tokens don't result in wrong reductions and possibly lose the best recovery state. The consequent increase in table size is quite moderate.
* The "typical case" application exposed via the runtime module now includes error reporting.

At this time, it should be quite feasible to use the deterministic scan and parse engines in production.
The APIs for those bits are about settled.

Usability in real projects (typically DSLs) is now quite decent in deterministic mode. Minimal-LR1 has had some shakedown and common abstractions with Canonical are exploited. Decent-looking error reporting in context is now easier. Table generation and compaction performance is actually pretty good. GLR remains incomplete. Seeking more example applications (beyond 3GLs) to bundle complete working real-world solutions.

This is now also very usable in a pre-compiled automaton mode and does a respectable job squashing these into minimal space while still having respectable algorithmic-complexity.

Downsides: the runtime could make error-response a bit easier, and GLR isn't finished yet.

Initial release to the general public of an MVP: Minimal Viable Product.