To what does the "we" refer in:

When we take the derivative of a constant Quantity, the result is a pure number, not a Quantity:

To the best of my knowledge, ever since Russell's abandonment of Relation Arithmetic in volume 4 of Mathematica (the original -- not the software) the foundation of mathematics does not even try to rigorously accommodate relationally dimensioned quantities. If there is scholarship in mathematics on this topic I would really like to see it because I basically blew my professional capital to hire Tom Etter to solve this problem -- critical to the foundation of programming languages (such as Mathematica ironically enough) at the largest DotCom era project: HP's "Internet Chapter 2".

That meta-problem aside, the example I provide has an additional "bug" evidenced by the final scalar output which, I would have thought, would be vector.

As for your last "bug":

ReleaseHold[aVec]

does what you don't expect because RealeaseHold only releases the outer Hold, but your aVec has two layers of Hold:

Hold[D[Hold[D[rVec, t]], t]]

which, after the release, becomes

D[Hold[D[rVec, t]], Quantity[tM, "Seconds"]]

The variable tM does not appear in the function, and the derivative is zero. rVec has no chance of being recognized as a vector made of lengths.

At least, this is my own interpretation at the moment. I am no expert of nested held expressions.