feat: further configuration of embedded ipfs daemon

[frrist]

Why

At present, the embedded IPFS node in bacalhau will listen on random ports. Users can configure the embedded IPFS node to listen on a "preferred" swarm address by setting the BACALHAU_PREFERRED_ADDRESS, but they are currently unable to configure which port the node listens on for swarm connections. The lack of determinism in setup makes it challenging to correctly configure firewall rules when deploying bacalhau to production-like settings, since the port the IPFS node listen on cannot be known at the time of deployment.

What

This PR allows the embedded IPFS nodes gateway, api, and swarm listening multiaddresses to be configured. Users can configure these values via the --ipfs-gateway-listen-addresses, --ipfs-api-listen-addresses, and --ipfs-swarm-listen-addresses flags on the serve command respectively. Alternatively, their correspond environment variables, or config file value may also be set.
To preserve backwards compatibility, users may continue to use the BACALHAU_PREFERRED_ADDRESS environment variable to configure the swarm address the embedded IPFS listens on, but should note this variable does not allow the port to be specified, it will still be randomly assigned.

Additionally, this PR makes changes to the behavior of bacalhau and its embedded IPFS node when the --ipfs-serve-path flag is set. The intent of the flag is to allow users to define a repo location of the embedded IPFS node to store content and its configuration. The current behavior deletes this repo when the bacalhau node shuts down. Resulting in any content stored in the repo being removed, and the identity of the embedded IPFS node being lost. The new behavior will preserve the content of the embedded IPFS nodes repo across bacalhau restarts, maintaining any data the embedded IPFS node stored as well as its identity. If the --ipfs-serve-path flag is not set the behavior remains unchanged - the repo is considered ephemeral and removed when bacalhau shuts down.

Lastly, this PR adds an additional flag, --ipfs-profile, to configured the embedded IPFS nodes configuration profile. The default profiles remain flatfs when --private-internal-ipfs=false and test when --private-internal-ipfs=true.

[frrist]

A nice side affect of this change is that we can now interact with the embedded IPFS daemon via the ipfs binary. For example:

1. Start Bacalhau Server: bacalhau serve --ipfs-api-listen-addresses=/ip4/127.0.0.1/tcp/6001
2. Use IPFS binary: ipfs --api=/ip4/127.0.0.1/tcp/6001 id

[simonwo]

Looks like the config will end up the same if only the existing flags are used? So fine by me.

[rossjones]

Users can configure the embedded IPFS node to listen on a "preferred" swarm address by setting the BACALHAU_PREFERRED_ADDRESS

This is really just a work-around to how we choose addresses. Because we often choose to bind to 0.0.0.0 - this is a valid choice, but results in the service binding to all networks, whether they're 127, loop-local, private or public addresses. Then one of those addresses is chosen, and it isn't clear what the criteria for the choice is.

Ideally we should be more selective about what IP address we expect IPFS to bind to, using an internal network for private clusters, and a public one otherwise.

If we find ourselves having to set BACALHAU_PREFERRED_ADDRESS we would be better off just binding to the correct network in the first place.

[rossjones]

These two addresses are not the same. ::1 is the same address as 127.0.0.1, if we want to bind v6 to all v6 interfaces, we should use ::