tests: Use a static token location for random tokens

tests: Use static tokens with static locattions

tests: Use static token location for tokenizer test

tests: Fix test reference invalid name

tests: Fix test_tokenize not testing whitespace ...

This relied on draw_token_random which now doesn't include whitespace

tests: Document parse error exception codes

parse: Reserve keywords for reference names

tests: Move whitespace code to test_tokenize ...

The parser doesn't have to worry about whitespace, so it's not worth
worrying about it here.

parse: Parse a reference node

env: Fix reference to syntax.py

tests: Add garbled strategy for draw_token_random

src: Rename syntax to token

parse: Remove SyntaxType ...

Once again I find myself in the pattern of having a useless 'type' value
that just complicates my life. In this case the correct appproach is to
output the AST directly.

parse: Parse text and bools to AST types

parse: Check types when clearing notes, add extra tests

syntax: Refactor draw_syntax_random and add draw_syntax_texttoken ...

I realized draw_syntax_classified is the same as draw_syntax_random, so
just merge them.

Also add draw_syntax_texttoken explicitly meant to weed out cases where
token type is not set.

tests: Clean up syntax and parse code

tests: Clarify function documentation

build: Note about executable mode

build: Always zip creation system to Unix ...

This doesn't matter and it means we can have reproducible builds across
Windows and Linux.

build: Tidy write_file

build: Ignore hidden files

build: Replace Windows line endings with Unix ones

build: Open files as binary read-only

build: Add shebang to zipapps

build: Replace git date with 2000 ...

There's no real point having artifical differences in our zip files.

build: Rewrite to make a reproducible zip

Jenkinsfile: Only cleanup on success

Jenkinsfile: Run domut_ci before dotest_ci ...

domut_ci does light testing first and should catch errors earlier

env.ps1: Fix domut not working