Among other minor fixes, this allows the client to cache the URI swarm name
prefix, avoiding extra encoding every time it is needed.

To avoid having to build the whole name (which implies encoding the injector
public key) every time a URI swarm name is needed.

It was previously on some internal document.

This includes several fixes related with automatic swarm name computation:

  - Use lower-case, unpadded Base32-encoding for injector keys in swarm names.
  - Disable obsolete `announce-in-bep5-swarm` injector option in Docker
    wrapper script.
  - Show injector public key in the client front-end (to ease showing it to
    other users).
  - Also allow setting the injector public in Base32-encoded format for the
    `--cache-http-public-key` client option (to ease receiving it from other
    users in written form).

Padded or unpadded, upper or lower case.

Only explicit BEP5 swarm names conflict with automatic swarm name derivation.

The injector BEP5 swarm name is computed automatically now.

The injector BEP5 swarm name is computed automatically now.

This encoding is more amenable to human exchange shall we accept it as input,
so better start using it now and avoid having to change the general swarm name
derivation method further down the road.

Because whether something is to be stored is implied by the use of the
fresh_channel::injector_or_dcache route.

It adds no value and only causes additional traffic. It is because if
the injector_or_dcache access fails, it implies that connection to the
injector failed. And if that is the case, than neither the ::proxy
mechanism will succeed in connecting to it.

... injector's public key, protocol version number and resource's dht
group.

Also make dedicated functions to compute names of these (and injector
and uri) swarms.

Mainly so that a timeout on the origin and secure_origin routes don't
halt injector, proxy and dcache routes.

Not using optional<error_code>{}

Not based on how many the request router already attempted to start.
Because if one access is disabled, it shoulnd't delay the following
onces.

Instead of trying one route after another, start async jobs all at once,
but each next one with an increased timeout.

Previously if there was (e.g.) POST HTTP method used, these URLs would
use those routers.

Also changed the way how we force secure HTTPS over just HTTP.

To enable assigning more than one on_finish callback