Modules: Carry context from incoming requests to hooks

[scr-oath]

This change

1. Allows disconnects from the server to cancel further work
2. Uses the same timeout via Done channel rather than using (and leaking; never canceling) a time.After timer for every stage.
3. Recovers from panics, which would previously halt the app
4. Doesn't run hooks when the context is already failed (such as when client navigates away).

[scr-oath]

Fixes #3962

[bsardo]

Hi @scr-oath, can you please merge with master and resolve conflicts? Thanks!

[scr-oath]

Hi @scr-oath, can you please merge with master and resolve conflicts? Thanks!

DONE

[scr-oath]

FWIW, marking draft - we're fighting with some mobile SDK crashes related to the piece that adds the full-request timeout stuff… essentially, SDK crashes if the passthrough exists and we have a hook to remove it, which doesn't run if the entire request times out - the fix was fairly straightforward but grows this to have HoldAuction return the ctx.Err() so that that can be honored as a writeError call (not 200 OK).

This also has me thinking - that hooks should probably have some way of declaring that they are critical - as in - timeouts should be treated as reject (or whole-request timeout)

[scr-oath]

We discussed this a fair bit in the Prebid Server backlog review and it seems there are other concerns/needs that may need some thought/discussion.

1. Tmax should be used as a calculation for time to give bidders but not block the entire request
2. Some hooks should be guaranteed to run (if they time out, it should be treated as a reject)
3. If the client disconnects, some teams still want to run to completion to capture the information about the auction for their data processing.

Therefore, It would seem that a writeup and review of a solution should be done before implementation and PR is completed, which I'll tackle in an issue and/or google doc.