Add multi-step fitted value methods for TSLM and others

Reported at https://stackoverflow.com/q/70782047/144157

library(fable)
#> Loading required package: fabletools

model_fit <- tsibbledata::aus_production %>% 
  model(mod = TSLM(Beer ~ Gas + trend()))

fitted(model_fit, h = 2)
#> # A tsibble: 218 x 3 [1Q]
#> # Key:       .model [1]
#>    .model Quarter .fitted
#>    <chr>    <qtr>   <dbl>
#>  1 mod    1956 Q1      NA
#>  2 mod    1956 Q2      NA
#>  3 mod    1956 Q3      NA
#>  4 mod    1956 Q4      NA
#>  5 mod    1957 Q1      NA
#>  6 mod    1957 Q2      NA
#>  7 mod    1957 Q3      NA
#>  8 mod    1957 Q4      NA
#>  9 mod    1958 Q1      NA
#> 10 mod    1958 Q2      NA
#> # … with 208 more rows

^{Created on 2022-01-21 by the reprex package (v2.0.1)}

Start with the fitted(model_fit, h = 2) entry point and the TSLM example in the issue, then inspect how fitted values are implemented for TSLM and other models. Done means the multi-step call returns fitted values rather than the shown NAs, with behavior defined for the affected model methods.