Bug: file download time and throughput is not accurate

[tempusfrangit]

With the recent changes download timings are no longer accurate for time spent downloading an individual file in multifile mode.

This is related to some chunks blocking on connections so download_elapsed is effectively total_elapsed in all cases.

• Pget should have a way of communicating time spent for a chunk and aggregating it for a given file rather than download_elapsed including blocking time. This also breaks the throughput calculation.

• An alternative is to explore downloading one file at a time and allocating connections to meet the target chunk size before moving on to the next file.

If we go with communicating the download time up the stack, this feels like a legitimate use for values in the context (pointer to per-chunk data) instead of trying to pass data back up the whole stack.