One additional idea I did not mention there, but which I did mention a
while ago on this mailing list, is to have two variants of every
procedure for which both multiple vector arguments _and_ start+end
arguments make sense.  For instance, there would be a (VECTOR-MAP <f>
<vector> [<start> [<end>]]) and a (VECTORS-MAP <f> <vector> ...).
This has the disadvantage of lots of extra procedures, and still isn't
as good as a complete array slice API.  But in the absence thereof,
this may be a good solution, as it provides you with the best of both
worlds -- start+end everywhere versus multiple vectors in operations
for which they make sense & start+end where they do not --.  Thoughts?