Support getting async and sync handles for existing blob store items.

[rklaehn]

We should have a way to get a handle from the store that allows you to treat a blob as a file in a sync context (implement Read and Seek traits) or in a (tokio-flavoured) async context (tokio::io::AsyncSeek and tokio::io::AsyncRead).

[rklaehn]

There are a number of questions about this:

1. which version is the more important for people, the sync one or the async one? Personally I am not very fond of the local io traits from tokio, but I realize that lots of people use them.

2. does this need the out of process rpc, or not? We have spent a lot of work on an out of process rpc layer and on language bindings, but have come to the realization that lots of people use iroh directly from rust. So we are currently focusing on the "iroh is a rust library" use case most, and are deprioritizing cross process ffi and rpc. See https://www.iroh.computer/blog/ffi-updates

3. there are 2 ways to do this, both technically viable. Not sure if we should offer either of them or both, and which one to implement first. This is only relevant for partial blobs, but these will become more important as we move from sendme use cases to other use cases

• access only for verified data. Seeking to a place where the data is incomplete and then reading will produce an io error.
• access for all data. Seeking to a place where the data is incomplete will read zeroes.

I guess in the spirit of content addressing the first one would be the one to implement first. You only ever get data that is consistent with the hash you have opened, or io errors. However, the second version is much faster since it does not need to validate anything.

[maboesanman]

The API I would expect to find (which may or may not be realistic) is two functions:

read_completed(hash) -> Option<impl Read + Seek>

read_in_progress(hash) -> impl AsyncRead + AsyncSeek

The completed variants are basically pass throughs for file readers

The in progress variant would actually just wait until the bytes are available, indefinitely. Seeking around might inform the underlying blobs machinery to prioritize certain chunks. There may be other error conditions as well.

[maboesanman]

Perhaps the read_in_progress needs to be split into a seekable but unverified version and a non seekable verified version

[rklaehn]

Here is a very WIP PR that adds the ability to get an impl Read + Seek for a complete or incomplete file. It can also do tokio AsyncRead and AsyncSeek. What it does not do however is to trigger downloads on seek - that would be another thing.

#72