Browser pools: Reuse limited number of browsers (memory ⬇️, speed ⬆️)

[ftes]

Experimental browser pooling. Reuses browsers across test suites.
This limits memory usage and is useful when running feature tests together with regular tests
(high ExUnit max_cases concurrency such as the default: 2x number of CPU cores).

Pools are defined up front.
Browsers are launched lazily.

Closes #75

[ftes]

Also see non-pooled load tests results here:
ftes/phoenix_test_playwright_example#2

They compare favourably to playwright JS client.

[s3cur3]

I am very interested in this, not for any potential speedup, but because when I run a large test suite in CI, some of which is Playwright tests and some of which is normal unit tests, I frequently see timeouts within the PhoenixTest.Playwright GenServers because we failed to hear back from Playwright within 1.5x the specified timeout (not timeouts due to client code saying something like "click such-and-such button within 2 seconds"). It'd be nice to be able to run the full test suite with max parallelism, while limiting the number of concurrent Playwright browsers to try to make sure they're not resource-starved.

[Valian]

I am very interested in this, not for any potential speedup, but because when I run a large test suite in CI, some of which is Playwright tests and some of which is normal unit tests, I frequently see timeouts within the PhoenixTest.Playwright GenServers because we failed to hear back from Playwright within 1.5x the specified timeout (not timeouts due to client code saying something like "click such-and-such button within 2 seconds"). It'd be nice to be able to run the full test suite with max parallelism, while limiting the number of concurrent Playwright browsers to try to make sure they're not resource-starved.

Exactly the reason why I made #75 👍🏻 👍🏻

[ftes]

I've pushed a v2 (curated the initial AI implementation).
Now, browser pooling is opt in. Pools are explicitly defined up front for full control.

Before merging and releasing, would you care to test it in your projects? @s3cur3 @Valian
See usage docs in PhoenixTest.Playwright.BrowserPool.

[ftes]

Merging & releasing. Since it's opt-in, no need to worry in case there are problems.
Still very happy to hear any feedback.

[s3cur3]

@ftes This is fantastic! Thank you so much. Yeah, I've dropped the browser pool size to half the schedulers_online and the overall reliability seems much better.