testharness: Use the correct lastStartTag value when running tests with “Script data state” as the initial state

[sideshowbarker]

This change ensures that TokenizerTester uses the correct, expected lastStartTag value from the test source when the initialStates value in the test source includes Script data state — or else if the test source specifies no lastStartTag, then script is used as the value.

Otherwise, without this change, TokenizerTester is hardcoded to always use script as the lastStartTag when running a test case with “Script data state” as the initial state. That discrepancy causes us to fail the following case from the html5lib-tests suite:

• https://github.com/html5lib/html5lib-tests/blob/master/tokenizer/domjs.test#L219-L223 (“lowercase endtags”)

[hsivonen]

I pushed this via the command line as f23387a8e09711a2a458bc8f4ea2eaec5ddb8fd8 and now I don't know how to make the GitHub PR metadata say merged instead of closed.

Thanks and sorry about the bad metadata state.