Name columns from named non-symbol arguments in xnew_data()

[joethorley]

Closes #109.

Note

Stacked on #102, which this builds on directly.
Base is fix-99-named-symbol-column; retarget to main once #102 merges.

#102 applied the argument name to the value only for bare symbols.
Every other named argument reached tidyr::expand() with its name intact, which gave two
bugs:

d <- tibble::tibble(a = 1:5 + 0.5, b = factor(letters[1:5]))

xnew_data(d, z = new_seq(b, .length_out = 2))$z   # bare vector
#> before: a b c d e      after: a b

names(xnew_data(d, z = xnew_seq(b, .length_out = 2)))   # one column tibble
#> before: "a" "b" "z"  where z is z$b       after: "a" "b" "z"  where z is a factor

The first is the idiom in xnew_seq()'s own examples, so it was reachable from the
documentation.

Approach

expand() sorts and deduplicates an argument's values whether it is a bare vector or a
data frame, so that is not the distinction.
The one behaviour that differs is factors: a bare factor vector is expanded to all of its
levels, whereas a factor held in a data frame keeps only the values present.
That level expansion is what discarded .length_out and .obs_only.

xnew_column() therefore gives every named argument the treatment #102 gave named symbols:

• a bare vector is wrapped in a one column tibble named after the argument;
• a one column data frame has that column renamed;
• a multi column data frame is packed into a named data frame column, exactly as today.

The dispatch is on the value rather than the expression because the type is not knowable
statically, so the helper is inlined into the quosure -- it is internal, and the quosure is
evaluated in the caller's environment where it would not otherwise be in scope.

The outer name is now dropped for every named argument rather than only for symbols, which
also removes the map_lgl() scan of the quosures.

Checks

• R CMD check: 0 errors, 0 warnings, 0 notes.
• Full test suite passes with no snapshot changes; the existing named-argument snapshots
  (b = 8:10, z = "zed", a = new_value(a)) are unaffected because those values are
  already sorted and unique.
• New tests cover the bare vector, one column data frame and multi column data frame cases.