tagGetAttribute(x, attr) returns a list if one of the values is not atomic

[schloerke]

Could not approach by flattening and splitting values. Too many downstream dependencies reach into the attribs.

Fixes #205

To get around more complicated attribs values, a list will be returned if any value is not atomic. We still recommend that everyone use character values only (or values that can be pasted together).

[wch]

Following up a discussion we had: We decided that we would always collapse multiple entries with the same name into a single vector/list, allowing type promotion to happen normally.

For example, div(x = "a b", x = "c") would result in an attribs data structure of c("a b", "c").

If the input includes a list or other complex data structure, like div(x = c("a", "b"), x = span("hello")), the attribs data structure would be list(c("a", "b"), span("hello")). (Note that this would generally be a bad idea to use!)

Note that we also need to be a bit careful about factors.

An additional thought I had afterward: we could add a parameter named split to tagGetAttribute() to do the following:

z <- div(x = "a b", x = "c")

tagGetAttribute(z, "x")
#> [1] "a b c"

tagGetAttribute(z, "x", split = TRUE)
#> [1] "a" "b" "c"

This would make it more convenient to check if the attribute contains a specific value, like "b".

I'm not sure the best way to handle attributes which are lists here, though.

[schloerke]

After more discussion, users are already abusing x$attribs$class. To change that structure would be very problematic.

---

With the latest implementation, characters are returned as normal.

z <- div(a = "x", a = "y")

z
#> <div a="x y"></div>

tagGetAttribute(z, "a")
#> [1] "x y"

---

But non atomic values are returned using a list structure, avoiding a call to as.character().

w <- div(a = "x", a = list(obj = TRUE))

tagGetAttribute(w, "a")
#> [[1]]
#> [1] "x"
#> 
#> [[2]]
#> [[2]]$obj
#> [1] TRUE

w
#> <div a="x TRUE"></div>

[schloerke]

I think this PR is ready to be merged.

The comment above shows a typical use case and a not-recommended use case.

The internal attrib structures are not altered and are only adjusted when printed or retrieved with tagGetAttribute()

[cpsievert]

LGTM pending minor suggestions