spec: make the ABNF read the sentences ergo.rego reads

[jbpros]

Follows #241. Fixes every disagreement between spec/sentences.abnf and ergo.rego that its check finds, and empties the list in spec/OPEN.md.

#241's check missed most of the cases. It only collected sentences under assert, for and from, not checks written as c: "x is 1", so it read 79 case sentences instead of 324. This PR collects those too, and that turned up three more groups, fixed here as well.

ABNF

• Spaces inside keywords. Keywords of several words, like is not empty, if any, and named by, now take WS between their words, so x is not empty parses. sentences.md:212 already said a run of spaces reads as one.
• Spaces at the ends. The ABNF says spaces at either end of a line are dropped before it's read, as ergo.rego does.
• Elapsed time. It can now go in a whose, after a plus, and alone on the right of a phrase, like vulns whose days elapsed from found to now is more than 30 or n is days elapsed from started to ended. ergo.rego reads these and reports an elapsed time isn't supported yet except alone before a phrase, so a parser built from the ABNF reads the same sentences. spec: say what the forms ergo.rego doesn't run yet mean and list them as known differences #243 takes this further and says what these forms mean.
• Lists without spaces. x is one of "a","b" parses, as sentences.md:212 says a list can be written with or without a space after each comma.

check_grammar.py

• It reads every check in the cases.
• It applies [sentence.words] after the ABNF parses a line, as sentences.md:407 says a parser has to. So count of approvers is empty doesn't parse as is followed by a field called empty. The word list is read from that rule in sentences.md, not copied into the script.
• It drops spaces at either end of a line before the ABNF parser reads it.

check_grammar.py now reads 6565 sentences with 0 disagreements, and still 0 with #243 on top. check_rules.py passes. No .rego changed.

🤖 Generated with Claude Code