It now guards against reading infinite files such as `/dev/zero`.
And most readings are bound with a (lax) limit.
As a side effect, more checking are done when reading a file and
overall the reading procedure is more robust.
PR #19095.
Previously we need a file buffer that is as large as the file size and
this could be a problem when user has less free memory available or
having very large data. Now with the help of `FileOutputIterator`,
we can have a much smaller, fixed size immediate file buffer and also
the code looks nice with `lt::bencode()`.
