NBD handle fatal errors in async read/write operations

[jimmyaxod]

silo/pkg/storage/expose/nbd_dispatch.go

Line 13 in fd58200

// TODO: Context, and handle fatal errors

The fatal channel is used here

silo/pkg/storage/expose/nbd_dispatch.go

Line 352 in fd58200

d.fatal <- err

AND

silo/pkg/storage/expose/nbd_dispatch.go

Line 307 in fd58200

d.fatal <- err

It is only used when executing asynchronous reads and writes, if it can't send the response, the error gets put in this channel.

The main Handle() dispatch loop should check the fatal channel, and if there was an error, return from Handle() with that error.
eg insert a select here

silo/pkg/storage/expose/nbd_dispatch.go

Line 153 in fd58200

// fmt.Printf("Read from [%d in, %d out] %v\n", d.packets_in, d.packets_out, d.fp)

[jimmyaxod]

We should also change this:

silo/pkg/storage/expose/nbd.go

Line 120 in fd58200

_ = d.Handle()

To not ignore errors from handlers, then we'll need to do something with that error.... what? Log it would be a start.