Drop all rows with empty slice() calls

[DavisVaughan]

This revisits a point brought up in #2252

I went to remove the early exit in slice_rows() and discovered it was actually being used to ensure that slice() with no ... returns all rows. I think this is theoretically inconsistent.

I think that if we have:

df <- data.frame(x = 1:2)

then:

filter(df)

# Starts with this slice index
c(TRUE, TRUE)

# Uses `&` to combine it with the locations specified by `...`

# So it returns all rows

and:

slice(df)

# Starts with this slice index
integer()

# Uses `c()` to combine it with the locations specified by `...`

# So it returns no rows

This is inline with the logic mentioned here:
#2252 (comment)

And with these two examples:

df <- data.frame(x = 1:2)

slice(df, c())
#> [1] x
#> <0 rows> (or 0-length row.names)

slice(df, 1, 1)
#>   x
#> 1 1
#> 2 1

And it allows us to remove the early exit, which to me has always been an indication that the theoretical logic is sound

"e53f5c67-a3fa-4b29-93b0-4df2710f40fa"

[DavisVaughan]

I took a look through the 137 broken revdeps, and the only one that looks related to this is rsample, where I actually added some overly strict tests. I will go fix that, otherwise I think this is fine to merge

rsample pr: tidymodels/rsample#380