Tries cleanup

[lorban]

Replace all Trie usages in the code base with a more abstract hierarchy of *Index interfaces, and do not directly expose the implementations, requiring the use of a factory exposed via the builder pattern to create instances.

This should help maintenance as the implementation will be chosen based on a set of requirements (case-sensitivity, mutability...) that the factory can interpret as it wants to provide the best implementation complying with them.

Clears the path for a proper fix for #5291

[lorban]

I've now removed all references to the Trie interface and its implementations across the code base. Three things are now left, in order:

1. Fix some bugs in ArrayTrie (there seems to be some off-by-1 errors making the tests fail).
2. Finalize the *Index interfaces.
3. Seriously ponder the Builder interfaces, maybe merge them all into a single one. The content initializing methods (with, withAll) and the config methods (caseSensitive(), fixedCapacity()) deserve a good dose of brain juice, especially the latter's defaults.
4. Clean up the implementations.

[lorban]

@lachlan-roberts I've added you as a reviewer to get your opinion about the look and feel of the Index, Index.Builder and MutableIndex API. Don't bother with the implementations as they're just a big hack at the moment so please just focus on how elegant or clunky it is to create and use indices.

[gregw]

Looking good!
Can we push on this to get it in before a .0.0
Even if the impls are not perfect, they are the same as we had before and the Builder abstraction would be good to get in for a .0.0

[gregw]

Hmmmm not having isFull changes this a bit... plus I;m not sure it is right. Maybe something like:

Suggested change

	if (_field == null)
	{
	HttpField field = new HttpField(_header, caseInsensitiveHeader(_headerString, _header.asString()), _valueString);
	if (_fieldCache.put(field))
	_field = field;
	}
	else
	if (_field == null)
	_field = new HttpField(_header, caseInsensitiveHeader(_headerString, _header.asString()), _valueString);
	if (!_fieldCache.put(field))
	{
	_fieldCache.clear();
	_fieldCache.put(field);
	}

[lorban]

Ideally, we should rely on some eviction algorithm in the index instead of clearing it. But that's quite a can of worm that should be left closed if possible.

Let's use that if (!put) clear & put logic, it should be roughly equivalent to what the old code is doing.

[gregw]

I think before it just filled up the cache and left it as is.... so maybe we just put and ignore the return for now.... we can refine in future PRs

[gregw]

No show stoppers for me now... but leave the issue open so we can refine a bit after .0.0