Hyper 1.x upgrade

[Niel Drummond]

The recent release of hyper 1.0 and onwards has changed a large portion of the API that Bollard was exposed to. I'll try to keep this ticket up-to-date with my findings.

This piece of work should require these independent parts, where I'm happy to have some help if there's motivation.

• Upgrade hyperlocal, which is possibly no longer maintained. It looks like a fairly straightforward upgrade if using the hyper-util legacy client, at least on the client-side, but that might be undesirable long term. (any help here updating / adopting much appreciated).
  • There is an existing PR that works for Bollard's needs.
• Upgrade hyper-rustls - there is a PR being worked on already, so I'm hopeful this work should be useable soon, and it looks useful as a reference.
  • This aforementioned PR is merged and release 0.26 contains the needed hyper upgrade.
• Hyper 1.x uses significantly different semantics of Request/Response body types, streaming and upgraded connections, which I've been working on in Bollard in a local branch.
  • Hyper 1.x has extracted the async read/write traits in order to distance itself from the Tokio type system and let other execution runtimes such as Fuchsia / Curl be used.. Unfortunately, in Bollard we still uses these traits to frame payloads, and the buildkit feature uses it for grpc communication, so we'll need to provide tokio wrappers.
• There is a "named-pipe" module in Bollard, which I haven't looked at yet, that would need to be upgraded to use Hyper's new API - I'm wondering if this should be replaced by something else or we can break it out into another crate. Any help here much appreciated.
• In parallel the rustls 0.22 upgrade is possibly also needed, which does look like a smaller change in the API, but nevertheless could be tackled separately.
• yup-hyper-mock also needs to be upgraded to use hyper 1.x