Support pull diagnostics and use them for testing

[jneem]

Push diagnostics (where the LSP decides when to publish, and sends notifications) are hard to test reliably, because the harness doesn't know when and how many diagnostics to expect.

This PR adds support for "pull" diagnostics, where there's a diagnostic request and the server responds to it. This simplifies the testing, and hopefully stops it from timing out at random.

This will probably conflict with the typechecker changes. I'm happy to wait and rebase.

[github-actions]

Bencher Report

Branch	more-reliable-nls-testing
Testbed	ubuntu-latest

Click to view all benchmark results

Benchmark	Latency	microseconds (µs)
fibonacci 10	📈 view plot 🚷 view threshold	386.41
foldl arrays 50	📈 view plot 🚷 view threshold	1,571.70
foldl arrays 500	📈 view plot 🚷 view threshold	5,818.90
foldr strings 50	📈 view plot 🚷 view threshold	5,834.10
foldr strings 500	📈 view plot 🚷 view threshold	50,933.00
generate normal 250	📈 view plot 🚷 view threshold	46,578.00
generate normal 50	📈 view plot 🚷 view threshold	1,876.80
generate normal unchecked 1000	📈 view plot 🚷 view threshold	2,833.80
generate normal unchecked 200	📈 view plot 🚷 view threshold	653.43
pidigits 100	📈 view plot 🚷 view threshold	2,756.20
pipe normal 20	📈 view plot 🚷 view threshold	1,277.90
pipe normal 200	📈 view plot 🚷 view threshold	8,479.20
product 30	📈 view plot 🚷 view threshold	747.47
scalar 10	📈 view plot 🚷 view threshold	1,312.10
sum 30	📈 view plot 🚷 view threshold	746.70

🐰 View full continuous benchmarking report in Bencher