[WIP] feat: replace uwsgi with granian

[mlabeeb03]

PLEASE REACH OUT TO @ahmed-arb FOR UPDATES ON THIS PR.

closes #937

This PR replaces uwsgi with granian.

All uwsgi related docs have been updated.
All uwsgi related patches and configs values have been renamed.
All uwsgi related files have been deleted and replaced with granian equivalents.

Here are the uwsgi configs/features we are currently using and their granian equivalents.

• static-map = /static=/openedx/staticfiles/

• static-map = /media=/openedx/media/
  Even though --static-path-route & --static-path-mount are available, they can only server from one path, in order to server both of them we must use caddy.

• http = 0.0.0.0:8000
  --host & --port

• buffer-size = 8192
  --http1-buffer-size 8192

• wsgi-file = $(SERVICE_VARIANT)/wsgi.py
  Use this parameter $SERVICE_VARIANT.wsgi:application

• processes = $(UWSGI_WORKERS)
  --workers

• ⚠️ thunder-lock = true
  Apparently granian does not have thundering herd problem, see Related Granian discussion

• single-interpreter = true

• enable-threads = true
  Threads are enabled by default. You can explicitly set them with --runtime-mode. Set the number of threads with --runtime-threads.

• http-keepalive = 1
  --http1-keep-alive enabled by default

• ❌ add-header = Connection: Keep-Alive
  Can not add arbitrary headers with granian, does this affect us if we have --http1-keep-alive?

• die-on-term = true
  Granian handles SIGTERM as a shutdown signal by default

• lazy-apps = false

• need-app = true

• no-defer-accept = true

• master = true

• py-call-osafterfork = true

• vacuum = true

• Reload
  We can define a file in the --reload-paths and changing that file will restart the server. We also need to install the reload dependency and also unable the --reload config.

[ormsbee]

Just a warning that I have a suspicion that MIT's switch to granian may have been responsible for exposing a multi-threading related bug in grading. This wouldn't be any fault of granian's, but more a case where granian's deployment strategy exposes already-existing problems. This is highly speculative on my part, and the investigation is ongoing. I just wanted to bring this to your attention early. The bug needs to be fixed regardless, so I don't expect it to block granian in the long term, even if this hypothesis turns out to be true.

[ormsbee]

This is for watching a directory and reloading if there are file changes detected, right? Just out of curiosity, why enable this in a prod environment?

[mlabeeb03]

We had a reload-uwsgi command as well. Here is the rational behind doing it. This is just a replacement for that. Now the reload-granian command puts current timestamp in the file mentioned in --reload-paths to trigger a reload.

[ormsbee]

Quick update: We believe the multi-threading issue has been fixed, though we're still waiting for MIT to deploy it on their instance next week to verify that it resolves the issues they've been seeing there.