tslm rewrite, datamat helper function and forecast.lm improvements

tslm rewrite
- Support for mts objects and mlm fitting
- More robust data object finding
- More validity checks
- Converts/renames function() objects so that they can be forecasted
- Allows trend+season value overrides
- More diverse subset operations and uses, also attempts to re-create ts
attributes of subsetted objects
- Allows `[` subsetting of variables and all other special/primitive
functions (e.g. mean())
- Stores data in 'model.frame'-esque manner, so ts/matrix attributes are
stored within data.frames

datamat helper function
- Retains matrix and other information within data.frame objects
- Renames formula() objects within dataset for tslm and forecast.lm
- Looks recursively within data.frame objects for source columns
(flattening)

Forecast.lm
- Can now forecast functions() in models when used in combination with
tslm
- Can now override trend + season for user specified values in newdata