workload: avoid some string format/parse roundtrips

[dt]

Previously we were formatting UUIDs and Timestamps into strings, and allocating those strings, to put them into the column vector only for import's workload reader to parse the strings back into native types as it read the vector, accounting for 5% or more of total CPU usage when IMPORT'ing from workload-generated TPCC data. Keeping these two types in native representations through the vector avoids this.

[cockroach-teamcity]

This change is 

[yuzefovich]

but there are a couple of test failures.

Reviewed 1 of 1 files at r1, 3 of 3 files at r2, 2 of 2 files at r3, all commit messages.
Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @dt)

[dt]

I spent a couple hours on the failures and it is kinda annoying: we don't have the desired sql types in the tests and only have the collapsed colVec types, so there is no way to tell what is a string from bytes from uuids since they are all colvec bytes. In real crdb usage we have the destination type to know how to parse those bytes but we don't in the tests.

I'm gonna put this on ice until after PTO / indefinitely (though it is big chunk of workload import cpu time) until I can figure out these silly tests.

[yuzefovich]

First two commits should pass CI, right? Maybe just merge those two for now then?