Consider deferring attached to emulate 1.0 callback timing

[kevinpschaaf]

From @sorvell on August 29, 2016 21:27

Related issues: #71, #96.

In 1.0, elements were ensured to be readied before being attached. This allowed use of local dom and (at lease naive) measurement in attached.

This was achieved specifically by deferring attached until after ready if the system tried to call it too soon.

We could retain this guarantee without explicitly deferring attached by readying (flushing) an element before attaching its local dom. This would change the guarantees around ready. Specifically, it would be unsupported to look at shadowRoot in ready.

Perhaps we should simply add back the defer attached behavior from 1.0 to ensure as little breakage for users of the 'legacy' api as possible.

Copied from original issue: PolymerLabs/alacarte#102

[sorvell]

This will likely not be addressed in Polymer.Element where we are not inclined to introduce new lifecycle (attached). Based on feedback we'll see if we need to add guidance around how to account for this.

[kevinpschaaf]

Resolved via #3947