As mentioned by the OP in the comments, Reviel Netz is an academic who thinks about these kinds of questions.  For example, he has a great chapter in "The Archimedes Codex" (by Netz and his coauthor William Noel) on the topic of how the Greeks labelled their figures.  He explains how the critical method - comparing and contrasting different manuscripts to reconstruct a common origin manuscript - can be applied to the various figures in the extant codices of the Archimedean tradition.  It's debatable how close the critical method can get to constructing what the figures drawn on the beaches of Syracuse in the third century BCE might have looked like.

However, Netz also notes the innovations of the medieval scribes.  He states (p. 111):

> Medieval scribes, at this point, took some crucial steps in the invention of more effective typesetting interfaces.  The use of several cases - an uppercase alongside a lower case - is of great value.  It allows us to separate the letters referring to the diagram (which remains in upper case) from the rest (which is now in lower case).

Thus, Netz appears to state that the *real* innovation is not so much writing in majuscule for the figures, but in writing a mix of majuscule and miniscule at the same time.  As pointed out by others, this is a medieval construct.  

(Netz states in passing that the ancient Chinese used different colors to identify the components of their figures.)