Add support for patterns(cols=user_provided)

[tdhock]

Closes #6498

Before this PR:

> melt(data.table(x1=1,x2=2,y1=3,y2=4),measure.vars=patterns("2",cols=c("y1","y2")))
      x1    y1    y2 variable value
   <num> <num> <num>   <fctr> <num>
1:     1     3     4       x2     2

above result is unexpected since variable=x2 which is not present in cols.

After this PR:

> melt(data.table(x1=1,x2=2,y1=3,y2=4),measure.vars=patterns("2",cols=c("y1","y2")))
      x1    x2    y1 variable value
   <num> <num> <num>   <fctr> <num>
1:     1     2     3       y2     4

above result is better since y2 variable is present in cols.

this provides functionality similar to #6077

[tdhock]

on my local windows machine, all tests pass.
this could be a "breaking change" since patterns now outputs character instead of integer.
But when patterns() is used in the context of melt() or .SDcols, it gives the same result before/after this PR. (so no breakage for the typical use case)
And I don't think people should be using patterns() otherwise (although it is exported).

[hongyuanjia]

Thanks @tdhock. I love this change. For me, and I believe most DT users, the primary use of patterns is to extract columns in melt and .SDcols, so no break changes in this aspect.

[github-actions]

Generated via commit d3bea19

Download link for the artifact containing the test results: ↓ atime-results.zip

Time taken to finish the standard R installation steps: 3 minutes and 13 seconds

Time taken to run atime::atime_pkg on the tests: 9 minutes and 21 seconds

[Anirban166]

LGTM!