[racket] Should we _always_ use `only-in` or `prefix-in` when `require`-ing packages?

I did "only-in" for a while in a large system, proactively, for some of 
the reasons you mention.  I later decided it was a tedious and 
ugly-looking pain in the butt with no actual benefit shown after years 
in practice.

I currently try to have reusable modules provide names that are 
reasonably unlikely to conflict, and calling that good enough.  This 
includes trying to put the package name in each provided identifier 
somehow, perhaps in an English phrasing way, to hopefully give 
programmers a better idea of where the identifier is coming from, as 
well as to reduce conflicts.  
("http://www.neilvandyke.org/racket-roomba/" is not the absolutely best 
example, since some of the names are based on opcodes from the spec, but 
it's a recent open source example.)

I occasionally see someone doing a reusable Racket module that provides 
generic names (say "get" or "open"), as if it's designed to be used with 
"prefix-in" or in small programs that don't use many modules.  It can 
give simpler-looking demos of example code (e.g., code says lots of 
"move" and "turn" rather than lots of "move-robot2000" and 
"turn-robot2000"), but I think it turns out to be tedious (now someone 
has to do "prefix-in robot:" anyway, or the module doesn't play well 
with other modules.  It also doesn't play well with Racket documentation 
search, which likes us to have unique names even though modules mean we 
don't have to.  It also is not great for debugging, when you have only 
an error-message string like "get: could not get" and it's coming from 
an indirectly-required module that you didn't even know was being used.

Neil V.