Enhancement: support larger file download

[anotherjesse]

@daanelson has shared that https://storage.googleapis.com/replicate-weights/llama-13b-fp16.tensors which is ~24GB.

To compare, gcloud can download this in parallel between 1-2 GBps

[joehoover]

I just have an anecdotal sample size, but I've found that pget works as-is for that model when run on a A100 instance.

Download time with pget was between 21-24 seconds. I tried tweaking -c and found that -c 10 to -c 12 seemed to slightly improve on speed obtained with the default.

Tests with gcloud yielded downloads between 16-24 seconds (with download speeds ranging from 1.1-1.7 GBS).

[tempusfrangit]

Potentially it makes sense to also compare to available ram, if one cannot buffer the whole file into memory use scratchspace and bind files together after.

[tempusfrangit]

Partially covered by #177