Implement load balancing of partitions among a miner's deadlines

[anorth]

Follow-on work from the initial Window PoSt implementation is to assign new sectors to deadlines so as to balance the load for a single miner between its deadlines.

A simple sort (as suggested in #402) is attractive, but not correct if there are a larger number of partitions to be filled than deadlines of equally-minimal weight. The full solution probably involves building a heap or maintaining a sorted list in order to repeatedly add the next partition to a minimal deadline, taking into account partitions just added.

See also #402 and #432.

[anorth]

I've broken the randomisation out to #432, as it is lower priority than simply load-balancing. I think that means that the fix for #402 will resolve this too. cc @ZenGround0

[anorth]

Quick estimates: for a miner to onboard 1EiB of storage in 1 year with 32GiB sectors, they'll be filling about 40 partitions per proving period. This puts it in the range where deciding between the O(n^2) maintenance of a sorted list vs O(n log n) but more complicated maintenance of a heap isn't clear. The heap is overkill for miners onboarding at half that rate.

64GiB sectors will halve that, so maybe go with the simple sorted list (example).