Ensure dfm_lookup returns dfm with all the keys

[koheiw]

This is a major bug. dfm_lookup() returns a dfm without columns when there is no matches.

> toks <- tokens(c("aaaa aaaa", "bbbb"))
> dfmt <- dfm(toks)
> dfm(tokens_lookup(toks, dict)) # Correct
Document-feature matrix of: 2 documents, 2 features (100.0% sparse).
       features
docs    key1 key2
  text1    0    0
  text2    0    0
> dfm_lookup(dfmt, dict) # Not (not features)
Document-feature matrix of: 2 documents, 0 features.

[codecov]

Codecov Report

Merging #1946 into master will decrease coverage by 0.01%.
The diff coverage is 50.00%.

@@            Coverage Diff             @@
##           master    #1946      +/-   ##
==========================================
- Coverage   91.27%   91.26%   -0.02%     
==========================================
  Files         100      100              
  Lines        6318     6319       +1     
==========================================
  Hits         5767     5767              
- Misses        551      552       +1     

[kbenoit]

Interestingly I noticed this too today, when working on the sentiment branch!

[koheiw]

I noticed this today when I was working on #1912