Add array streaming helpers

[arnetheduck]

An implementation of
#112 that introduces beginArray/endArray for the streaming creation of arrays.

In order to accomodate the need for intra-element plumbing, we add begin/end marker calls to the writeValue implementation which helps the writer keep track of each value being written and therefore allows it to inject the correct delimiters and indents.

Here's an example of writing a writeValue overload that writes an array nested in an object:

# Fully expanded form
proc writeValue(w: var JsonWriter, t: MyType) =
  w.beginArray()

  for i in 0 ..< t.children:
    writer.beginObject(MyType)

    writer.writeName("id")
    writer.writeValue(i)

    writer.writeName("name")
    writer.writeValue("item" & t.childName[i])

    writer.endObject()

  writer.endArray()

# Convenience form
proc writeValue(w: var JsonWriter, t: MyType) =
  w.writeArray:
    for i in 0 ..< t.children:
      writer.writeObject(MyType):
        writer.writeMember("id", i)
        writer.writeMember("name", "item" & t.childName[i])

Similar to the existing beginRecord/endRecord fields we add beginArray and endArray - we also take the opportunity to name beginObject according to its json-spec-derived name. The old name remains available.

This change introduces a compile-time backwards-compatibility break for custom writers that try to access the stream directly: they now have to delimit their value writing with w.streamElement(s): s.write ... where s is the stream variable. The block template enforces begin/end markers on behalf of the writer.

Further examples are available in the documentation.

With this change, we also deprecate workarounds like fieldWritten and endRecordField since a regular replacement exists in the form of consistent begin/end pairs.