Form.select_columns needs a way to distinguish structural dots from dots in the names of record fields

[jpivarski]

Version of Awkward Array

HEAD

Description and code to reproduce

Suppose you have an array like

>>> array = ak.Array([{"outer": {"crazy.dot": {"inner_a": 12.3, "inner_b": 123}}}])
>>> array.show(type=True)
type: 1 * {
    outer: {
        "crazy.dot": {
            inner_a: float64,
            inner_b: int64
        }
    }
}
[{outer: {'crazy.dot': {inner_a: 12.3, inner_b: 123}}}]

If you want to Form-select inner_a and not inner_b, you can't:

>>> array.layout.form.select_columns("outer.crazy.dot.inner_a")
RecordForm([], [])

When this problem is encountered in Spark, they use backticks. To do the same, we'd want

>>> array.layout.form.select_columns("outer.`crazy.dot`.inner_a")
RecordForm([], [])

to result in

RecordForm([RecordForm([RecordForm([NumpyForm('float64')], ['inner_a'])], ['crazy.dot'])], ['outer'])

which means a bit more parsing of the specifier.