Improve compression estimators for directory spooling
Once we have zstd compression in place, we should improve the estimated ratio in estimate_compression_ratio(). We could also do well to have that function consider the size of the input data, since bigger data sets compress better.
(I'm deliberately only saying zstd here, not lzma2: We should never do lzma2 compression on the fly, so we shouldn't need to estimate the size of the output for spooling purposes.)
Calling this "low" priority because "0.5" isn't that bad of an estimate.