perf: spawn tokio runtime once

[DaniPopes]

Motivation

We're spawning thousands of tokio runtimes which accounts for ~50-60% of the time spent in the main thread. These calls happen both in parallel (where there is almost no gain), and in the end clean up (which speeds it up significantly). This happens in RuntimeOrHandle::new().block_on(...) which spawns a new tokio runtime to run a single future, then drops it

Before (41% + 17%):

After (no recorded calls):

Solution

Make test functions async and create a runtime in the CLI dispatch

[mattsse]

omg, finally.

I have been dreading this for so long because this touches the test runner API.

tysm for doing this!!

this pulls in openssl again for apparently no reason, no idea why, perhaps the rustls feature? idk

[mattsse]

some failing tests, because of blocking calls I think

[DaniPopes]

Same as openssl, any clue how to fix? b325876

[mattsse]

the only cargo change I'm seeing is the rustls feature

https://github.com/foundry-rs/foundry/pull/5263/files#diff-d9d5adf55ead8df6e41955ab981380acb217885a608393d800515cb51a769c87

maybe try after reverting this, no idea why this would cause it though

[mattsse]

tysm!

step by step we're improving the test runner