Dont check in the grammar, but generate it at release time

[jneem]

This is an alternative to #2141, that avoids checking in a 22MB generated file. Here, we generate the grammar at cargo publish time. The main disadvantage of this one is that it's hard to test: although I've tested the individual commands, I haven't tested the release script as a whole because I didn't want to push changes to the stable branch.

I've simplified the build script so that it doesn't try to check the freshness of the generated grammar: it assumes that if there's a generated grammar already present then it's fresh. In principle the published crate could have an even simpler build process (with no build-dependency on lalrpop) but that would introduce more hard-to-test release stuff so I haven't gone that route.

[github-actions]

Bencher Report

Branch	dont-check-in-the-grammar
Testbed	ubuntu-latest

Click to view all benchmark results

Benchmark	Latency	microseconds (µs)
fibonacci 10	📈 view plot 🚷 view threshold	493.46
foldl arrays 50	📈 view plot 🚷 view threshold	1,829.70
foldl arrays 500	📈 view plot 🚷 view threshold	6,964.90
foldr strings 50	📈 view plot 🚷 view threshold	7,447.90
foldr strings 500	📈 view plot 🚷 view threshold	65,998.00
generate normal 250	📈 view plot 🚷 view threshold	48,437.00
generate normal 50	📈 view plot 🚷 view threshold	2,078.40
generate normal unchecked 1000	📈 view plot 🚷 view threshold	3,515.30
generate normal unchecked 200	📈 view plot 🚷 view threshold	795.53
pidigits 100	📈 view plot 🚷 view threshold	3,209.90
pipe normal 20	📈 view plot 🚷 view threshold	1,548.90
pipe normal 200	📈 view plot 🚷 view threshold	10,695.00
product 30	📈 view plot 🚷 view threshold	850.07
scalar 10	📈 view plot 🚷 view threshold	1,537.80
sum 30	📈 view plot 🚷 view threshold	839.84

🐰 View full continuous benchmarking report in Bencher