Implementing reader-side Get()s for ExpressionString derived variables

[eisenhauer]

No description provided.

[eisenhauer]

@lizdulac and @anagainaru This is a revival of the older PR. Main changes on the writer side are: 1) Like StatsOnly, it adds block-level metadata on the writer side and 2) Unlike StatsOnly, it doesn't actually do any of the allocations or calculations, just goes through the motion so that we get the same blocks info that you'd have generated if you did those derived variable calculations. On the reader side, we do the same stuff that we do for StatsOnly. That is, if someone asks us for a block of a derived variable that we don't have data for, we calculate the data and return it.

[anagainaru]

Will look at this tonight, it needs more attention than other PRs

[eisenhauer]

As of the most recent build, this was run with DerivedVariables enabled, so it does compile and run on all platforms. But master now has that turned off, so once updated or merged it'll be optional, not compiled or tested in CI. (Not sure what's going on with that one failure, but it isn't related to his PR.)

[anagainaru]

I would not propagate DoCompute all the way to the math functions. We can stop at the expression level, creating empty DerivedData (as I am suggesting above).

However, the more I think about this, the ExpressionOnly flag is on the entire Variable so I am not sure why you need to call ApplyExpression at all.

[anagainaru]

The tests pass with DerivedVariables ON. Greg feel free to revert if you don't like my changes. Otherwise this is ready to merge.

[eisenhauer]

Sorry, had to add one more thing to make sure that bpls isn't looking for min/max that doesn't exist for these vars. Approve again?