geda and pcb don't care if the pinnumbers are numbers or strings. As
long as they are the same.
From a preference point of view I like the pin numbers to match the
component data sheet.
This is the part I too would have thought was natural. I would have
thought
that "B", "C", and "E" were more sensible pin numbers in the symbols
than "1", "2", and "3".
Also, I have read the section about transistor pin mapping between
symbol and footprint here:
http://geda.seul.org/docs/current/tutorials/gsch2pcb/transistor-
guide.html
The author specifies two schemes for achieving this, and chooses one.
I would have thought a third scheme would be better, where all
transistor
symbols have pins labelled "B", "C", and "E", and there will be
different
versions of the TO92 (or TO5 or whatever) footprint, called "TO92-
EBC",
"TO92-CBE", and so on.