Integrate file-reading with streams

[gterzian]

Similar to #26743 (although perhaps smaller in scope and a bit easier).

When reading a file, we currently just loop and read chunks from the file "as fast as possible" and send each chunk to script over IPC.

servo/components/net/filemanager_thread.rs

Line 211 in f240140

pub fn fetch_file_in_chunks(

Instead, we could use a mechanism as described in the linked-to issue, where the stream in script would "pull" chunks of the file from net, based on it's own(and possibly user provided) queuing and high-watermark strategies.

That way, one could imagine a situation where a file is read from disk, chunks are pulled into script by a readable stream, then perhaps put through a transform stream of some kind, and then transmitted over the network as the body of a request(another stream). Then, as bytes of the request body would be transmitted over the network, this would result in more chunks being "pulled" from the file on disk(once the internal buffer of a stream would go below some desired size).

(One can also imagine a fast path where if it's just the native underlying source for the blob stream that is directly feeding data into another native underlying source for the stream of the request body, we wouldn't have to go through SpiderMonkey, and perhaps we wouldn't event have to send file data to script over IPC at all, although it might be hard to know in advance how the stream will end-up being used in script, if at all)

Currently we'd just have one thread n net reading the file as fast as possible, sending bytes to script over IPC where they would accumulate in the buffer of the blob stream.