Start axum migration

[syphar]

This starts the migration from iron to axum (see #1900 )

We're using a strangler-fig pattern, so we're running the old iron server in the background, which will be called when we don't find a route in axum. The service is based on adapted code from axum-strangler.

I migrated some handlers from web::sitemap and some static redirects, especially to show how I thing we should do:

• template rendering
• CSP
• error handling without ctry, cexpect or .unwrap/.expect
• request metrics recorder
• cache middleware

especially the combination of template rendering and CSP took some thinking, but IMO this is a good approach.

There is some duplication (impl_webpage + impl_axum_webpage, AxumNope + Nope, AxumErrorPage + ErrorPage), which was easier to do than figuring out how to design for both Iron & Axum in these structs. And since at some point Iron will be gone, we're fine IMO.

I tried to do most of the change in our TestFrontend so the tests just run as they are, so we're sure everything is fine.

other things that will come / are good to know

( see also master...syphar:docs.rs:new-web-axum )

• in web::routes -> a get_rustdoc will also inject a prefix-blocking middleware
• the old shared-resources handler will also be a middleware, if the rebuild isn't finished until then and we can replace it with a handler.
• trailing-slash handling is done by default in the axum router
• axum/tower graceful shutdown will be added when iron is gone

pending manual test

• do redirects include https, coming from cloudfront? ( see redirect_base)
• are request metrics recorded correctly?
• sentry error reporting / trace

which specifics are not covered with tests good enough?

[syphar]

cc @rust-lang/docs-rs for review, tests, inpout

[jyn514]

this is great :) I left a bunch of comments but nothing looks blocking.