Co-authored-by: Marcin Rataj <lidel@lidel.org>

This makes it easier to reason about.
Also added better helptext to each command explaining how the binary
data is encoded on the wire, and how to process it in userland.

Those output formats are undocumented and seem to be only used in tests.
This change removes their implementation and replaces it with error
message to use JSON instead.

I also refactored tests to test the --enc=json response format instead
of imaginary one, making tests more useful as they also act as
regression tests for HTTP RPC.

Testing against compatible version from
ipfs/go-ipfs-api#255

Making it clear what it does and why

This ensures `ipfs pubsub sub` returns the same peerids in the `From`
field as `ipfs pubsub peers`.

libp2p already uses base encoding, no need to double wrap or use custom
multibase.

We want to send payload in the body as multipart so users can use
existing tools like curl for publishing arbitrary bytes to a topic.

StringArg was created for "one message per line" use case, and if data
has `\n` or `\r\n` byte sequences, it will cause payload to be split. It
is not possible to undo this, because mentioned sequences are lost, so
we are not able to tell if it was `\n` or `\r\n`

We already avoid this problem in `block put` and `dht put` by reading
payload via FileArg which does not mangle binary data and send it as-is.
It feel like `pubsub pub` should be using it in the first place anyway,
so this commit replaces StringArg with FileArg.

This also closes ipfs#8454
and makes rpc in go-ipfs easier to code against.

Making sure we don't see regressions in the future.
Ref. ipfs#7939

See
ipfs/interop@344f692

- Adds HTTP RPC regression test that ensures topic is encoded as URL-safe
  multibase.
- Moves pubsub tests to live in unique range ./t032x

uses js-ipfs from
ipfs/js-ipfs#3922