Refactor tag resolve() API

[eemeli]

This is a breaking change for custom tags. The main thrust is to remove any link between the underlying CST implementation and tag implementations. That should make custom tags easier to write and understand (in particular for collections), and will allow for the CST implementation to be changed without further tag API changes.

Previously, custom tags used this sort of interface when parsing values:

import { YAMLSemanticError } from 'yaml/util'

function resolve(doc, node) {
  try {
    return magic(node.strValue)
  } catch (error) {
    doc.errors.push(new YAMLSemanticError(node, error.message))
    return null
  }
}

The new API looks like this:

function resolve(value, onError) {
  try {
    return magic(value)
  } catch (error) {
    onError(error.message)
    return null
  }
}

Here, value is either a string, YAMLMap or YAMLSeq, depending on the shape of its value. A tag should verify that it's getting an instance of its expected type. onError is a callback that takes the error message as its single argument and internally constructs a YAMLSemanticError from it.

Because mappings and sequences are now always initially resolved into their corresponding AST structures, the parseMap and parseSeq exports of 'yaml/util' are also dropped.