dfm_group() does not group documents when dfm is empty

[koheiw]

Describe the bug

require(quanteda)
#> Loading required package: quanteda
#> Package version: 3.2.5
#> Unicode version: 10.0
#> ICU version: 60.2
#> Parallel computing: 32 of 32 threads used.
#> See https://quanteda.io for tutorials and examples.
txt <- c("This is a text. Next test.")
corp <- corpus(txt) %>% 
    corpus_reshape()
toks <- tokens(corp)
dfm_group(dfm(toks))
#> Document-feature matrix of: 1 document, 7 features (0.00% sparse) and 0 docvars.
#>        features
#> docs    this is a text . next test
#>   text1    1  1 1    1 2    1    1
dfm_group(dfm(tokens_remove(toks, "*"))) # wrong
#> Document-feature matrix of: 2 documents, 0 features (0.00% sparse) and 0 docvars.
dfm_group(dfm_remove(dfm(toks), "*")) # wrong
#> Document-feature matrix of: 2 documents, 0 features (0.00% sparse) and 0 docvars.