Optimizition: use lm.fit instead of lm

[durszlaczek]

lm.fit is faster than lm. However, some statistics must be calculated separately (e.g. group means).

[pbiecek]

premature optimization is the root of all evil

[durszlaczek]

Agreed, just put it here not to forget about it. That's why there is no assignee for this issue.