Define db's associated upgrade tx to align with impls/tests

[inexorabletash]

Implementations and tests have create/deleteObjectStore throw "TransactionInactiveError" if the transaction has aborted but the abort event has not yet fired. The spec disagreed, or at least was imprecise about the definition of "finished".

To resolve this, databases are given an associated upgrade tx which is nulled out when the event fires, rather than relying on "finished". This simplifies text elsewhere too.

@brettz9 - can you review this change? This should address the outstanding issue you noted over in web-platform-tests/wpt#5612

Tracking bug for this is #192

[inexorabletash]

Testing for this is already covered in WPT:

IndexedDB/idbdatabase-createObjectStore-exception-order.htm
IndexedDB/idbdatabase-deleteObjectStore-exception-order.htm

[brettz9]

While I'm still checking (looks good so far), I did have the following concern/question... It is admittedly a concern about a missing test rather than spec issue, but I also wanted to confirm what the testing behavior would be if it existed...

idbtransaction-objectStore-finished.html, and idbobjectstore-index-finished.html cover the expected InvalidStateError throwing immediately after an abort (and suggest by their test names that "finished" at least definitively occurs within abort). However, I don't see any tests for an abort throwing immediately after another abort (as would be consistent with this understanding (of one subset) of "finished" since abort is to throw upon a transaction being "finished"). (FWIW, besides my manual checking, my regex to check the files was abort\([\s\S]*abort\(.)

Should there be such a test which checks for InvalidStateError immediately after a prior abort?

[brettz9]

Besides my somewhat-related question about the expected behavior of multiple aborts, the PR looks good.

[inexorabletash]

I don't see any tests for an abort throwing immediately after another abort (as would be consistent with this understanding (of one subset) of "finished" since abort is to throw upon a transaction being "finished").

Yes, it does look we're missing a test for that case.