Handle JS errors properly

[Philip Taylor]

Our current handling of SpiderMonkey errors is all broken: mostly we assume they don't happen, and if they do we report an error but don't propagate the exception. I've not noticed this being a significant problem in practice, but it's not nice.

I believe the way it's meant to work is (as ​here):

• If a JSAPI function returns JS_FALSE, it has reported an error and thrown an exception; either propagate JS_FALSE back into JSAPI, or use JS_ClearPendingException to catch it.
• To trigger a JSAPI error, use JS_ReportError and return JS_FALSE.