Should support naming the metric in OxQL queries, in addition to the special datum identifier

[bnaecker]

Today, one can write OxQL queries like get foo:bar | datum > 100, where datum is a "special identifier" that we cons up to mean "the value at each point in the resulting timeseries". This behavior is implicit, which isn't great, and also falls over the moment we do joins: which element of the resulting multi-dimensional value does that refer to? We currently say, "neither" and return an error.

We could instead let folks write an identifier that is the metric name itself. From the above example, that'd be get foo:bar | bar > 100. If we join multiple tables together, then you can still refer to each one separately in your query, assuming they have different names. We could also consider "fully-qualified" syntax like { get foo:bar; get baz:quux } | align mean_within(1h) | filter foo:bar > 100, which would let us completely disambiguate which one the filter should apply to.