JS: Performance tweaks in AMD, mayHaveStringValue, PropWrite

[asgerf]

• mayHaveStringValue used a recursion pattern through a super-call which the optimizer isn't handling very well. Switched to a more direct style.
• The above caused a bad join order in AmdModuleDefinition.getFactoryNode(), which can be avoided by caching AMD modules.
• It also caused bad join orders for uses of PropWrite.getRhs(), generally by putting TValueNode first in the join order. It turns out getRhs was actually recursive with hasPropertyWrite which I decided to fix, after which magic made it recursive again, which I fixed by adding pragma[nomagic] in a few key places.

Evaluations all show a speed-up of a few percent:

• Full/nightly
• XSS/big-apps
• security/smoke-test

Based on some quick tuple counting, the second commit, and third+fourth commit, independently cause a speed-up, whereas the first commit is dependent on the others.

[erik-krogh]

I like the mayHaveStringValue change 👍

I don't like using cached to fix a join-order, but I assume you explored other options.

The change in ObjectDefinePropertyAsPropWrite doesn't make the code prettier, but I get why it can make is faster.
We loose support for reflective calls to Object.defineProperty with this PR, but I think that is acceptable.

Nice speedup 🎉
LGTM 👍