[C++] Optimize LL1Analyzer

[jcking]

No description provided.

[jcking]

This optimizes LL1Analyzer by removing extra heap allocations, removing usage of shared_ptr where possible, and reducing stack usage by reducing the number of arguments passed on the stack recursively.

@mike-lischke

[jcking]

I also got rid of the finally usage because that's really only useful for exceptions and there is no code catching exceptions from LL1Analyzer, so a throw will kill th state anyway. I also used LL1AnalyzerImpl to avoid changing the LL1Analyzer contract, but I'm happy to modify LL1Analyzer directly if you prefer.

[mike-lischke]

That's tricky to review, as a large block was moved around and I cannot compare both. Also I don't see a real benefit of moving a part of the implementation into an inner class, but it cleans things up a bit, so I'm willing to approve, but pls check my comment yet.

[mike-lischke]

@parrt This patch can be merged.