RFC: chrome: use a dedicated Chrome instance

[freesteph]

Du commit:

We've had a long time problem on this project: the complete darkness surrounding the crawling done by Chrome within our processes. That's because we don't provide a browser URL to Ferrum, and so Ferrum boots its own Chrome process within its Ruby process1.

This means we're unable to correctly debug our jobs containers on Scalingo (we can't SH into our running containers, we cannot run ps on them), and that these containers inevitably grow very large because on top of their own overhead (our Ruby/Rails app) they have to allocate and run their own Chrome process which adds a fair amount of memory and CPU to their own.

Demonstrate a new approach here where we run Chrome separately, on a new Docker container we control (on Scalingo we can use a buildpack instead2) and then we just feed to Ferrum our local URL.

This allows us to log, inspect and monitor our Chrome business while keeping our Ruby job containers happy & safe, away from the nightmarish world of web browsers.

[freesteph]

@Josyann @yann120 effet de bord intéressant : l'erreur de la CI montre que la spec spec/models/checks/run_axe_on_homepage_spec.rb lance un navigateur parce qu'on mock mal le browser, ce qui explique pourquoi elle est particulièrement lente

[freesteph]

@Josyann @yann120 effet de bord intéressant : l'erreur de la CI montre que la spec spec/models/checks/run_axe_on_homepage_spec.rb lance un navigateur parce qu'on mock mal le browser, ce qui explique pourquoi elle est particulièrement lente

Pas du tout, au temps pour moi, c'est juste que ça mockait un peu dans le deep des fonctions bien au delà de la classe en question mais il y avait un garde-fou sur Ferrum::Browser qui empêche de lancer un navigateur.

#586