Thanks to Gabor Grothendieck for the following suggestions:

- Allow a formula interface:

backtest(ret.var ~ in.var + by.var, DF)

and maybe

backtest(ret.var ~ in.var | by.var, DF)

which would do a separate ret.var ~ in.var analysis for each level
of by.var.

- Allow backtest to still give full info, not just the spread (this
could be user specifiable) even if there is a by.var.  For example,
maybe the second formula above with a | could specify the full report.
