fix(tests): return actual download count in run_cli to catch missing output files

[snushev]

Closes #155. Removed the fallback that forced ok count to len(urls) when done == 0, preventing false positive test passes when no files are actually written.

[snushev]

My bad, noticed the mocked behavior

[LeyckerS]

@snushev — you spotted it yourself and closed it before anyone had to say a word, which is the right instinct and I would rather have that than a defended patch. No harm done at all.

For the record, since it is a genuinely tricky one: the stubs write no files, so len(os.listdir(out)) is 0 and run_cli substitutes len(urls). Removing the fallback alone leaves ok == 0 against len(urls) == 4, which turns a vacuous pass into a false failure. And it turns out there is a further wrinkle neither of us had: the same helper is used by the engine tests, where ok comes from engine.snapshot() and is real — so the assertion is dead through two callers and live through two others. I had that wrong in the issue too and have corrected it there.

#155 was already reserved for @XEDAB, who reported it — that is on my listings for not making it obvious enough, not on you.

If you would like something that is genuinely free right now, #83 is the most interesting one open: the interface offers one "streams" number for both providers while datanodes is hard-capped at 8 lanes, so the number on screen is not the number in effect — measured at roughly 40 of 48 slots busy for one provider against 4 for the other. #151 is smaller and more self-contained if you would rather. Say the word on either and it is yours.