Follow-up from "ensure read timeouts are recoverable"
As discussed here, it is possible in
netmgr_test for two TCP sends to occur before a read, which results in a read twice as long as expected. This is not currently causing a test failure because, as of !4930 (merged), we don't check for the block length to be exactly correct, we only check for a minimum size. This should probably be investigated, and we should either prevent double sends, or have the read callback loop over the block it reads and process all of the data.