Categorise methods and refactor code base

[PeteHaitch]

Should be possible to categorise methods based on a few criteria:

• Row-block processing vs. column-block processing vs. requires full matrix in-memory (at least under a naïve implementation)
• Endomorphic (returns matrix of same size as input) vs. returns a vector
• Other?

Can then greatly simplify codebase by writing code that handles each case and code that registers methods for each function in that category.

Furthermore, should now use DelayedArray::blockApply()/minfi:::blockApplyWithRealization()/minfi:::blockMapply()/minfi:::blockMapplyWithRealization() (moving the minfi functions over to DelayedMatrixStats, if not DelayedArray), along with minfi:::colBlock() and minfi:::rowBlock().

[PeteHaitch]

NOTE: Those in bold are implemented in DelayedArray.
NOTE: Consider registering a DelayedUnaryIsoOp or DelayedNaryIsoOp for 'matrix in, matrix out (endomorphic)' ops

Matrix in, vector out

• colAlls(), colAnyMissings(), colAnyNAs(), colAnys(), colCollapse(), colCounts(), colIQRDiffs(), colIQRs(), colLogSumExps(), colMadDiffs(), colMads(), colMaxs(), colMeans2(), colMedians(), colMins(), colOrderStats(), colProds(), colSdDiffs(), colSds(), colSums2(), colVarDiffs(), colVars(), colWeightedMads(), colWeightedMeans(), colWeightedMedians(), colWeightedSds(), colWeightedVars()

Matrix in, matrix out (not endomorphic)

• colAvgsPerRowSet(), colDiffs(), colQuantiles(), colRanges(), colRanks() (depending on value of preserveShape), colTabulates(),

Matrix in, matrix out (endomorphic)

• colCummaxs(), colCummins(), colCumprods(), colCumsums()