Security: improve exact index matching performance

[jaymode]

This commit improves the efficiency of exact index name matching by
separating exact matches from those that include wildcards or regular
expressions. Internally, exact matching is done using a HashSet instead
of adding the exact matches to the automata. For the wildcard and
regular expression matches, the underlying implementation has not
changed.

[elasticmachine]

Pinging @elastic/es-security

[tvernum]

LGTM

[tvernum]

The existing Automatons code throws an exception if a pattern starts with / but doesn't end with /.
So we could simplify all this checking down to a single if/else and still be compatible with existing behaviour.

if (indexPattern.startsWith("/") || indexPattern.contains("*") || indexPattern.contains("?")) {
    nonExactMatch.add(indexPattern);
} else {
    exactMatch.add(indexPattern);
}

I don't care strongly, but shorter and simpler seems nicer.

[jaymode]

++ thanks for noticing that. I misread the if statement in Automatons

[tvernum]

I think as a follow up, when we have more time, we should:

1. Centralise this behaviour, and use it whereever we do pattern matching. It might mean putting a new Patterns class in front of Automatons that handles this logic.
2. Do some benchmarking about whether "*xyz*" is more efficient to run as str.contains("xyz") (and similarly startsWith and endsWith) and include that into the hypothetical Patterns utility class. I suspect, without testing that we pay a high price for using Automaton for trivial patterns.