fix: propagate timeout errors from no-length body reader

[shreemaan-abhishek]

Problem

For a response without Content-Length, calling body_reader(max_chunk_size) enters the streaming branch of _body_reader.

If sock:receive(max_chunk_size) returns a non-closed error such as:

nil, "timeout", partial

the current implementation skips the special closed branch and then yields only str, which is nil:

local str, err, partial = sock:receive(max_chunk_size)
if not str and err == "closed" then
    co_yield(partial, err)
end

max_chunk_size = tonumber(co_yield(str) or default_chunk_size)

The caller therefore observes nil, nil, which is indistinguishable from a successful end of body. The original error and any partial data are discarded.

This behavior is present in both v0.2.3 and the current master branch.

Expected behavior

A timeout or other socket read failure should be returned to the caller rather than being reported as a successful end of body. The behavior for connection-close-delimited EOF should remain distinguishable so consumers can handle that framing correctly.

Reproduction shape

Use a response with no Content-Length and no chunked transfer encoding, send fewer bytes than the requested reader size, and keep the connection open beyond the configured read timeout:

local chunk, err = res.body_reader(8192)
-- Current:  chunk == nil, err == nil
-- Expected: err == "timeout"

A focused reader test with a socket stub returning nil, "timeout", "partial" should also reproduce the issue.