Parser refactor

[dgnorton]

The big change in this PR is how identifiers are scanned / parsed. The scanner used to treat "db"."rp".cpu as one identifer. It would be stored in a single string in the AST somewhere. The scanner API exposed SplitIdent and it was up to clients consuming the AST to split identifiers into segments, strip quotes from the segments, and figure out what each segment was (db?, rp?, measurement?). This PR simplifies things for client code by splitting identifiers and stripping quotes at the earliest stage (scanning) and changing the parser to store them in appropriate fields (db, rp, etc.) in the AST.

• Change Scanner.scanIdent() to scan "db"."rp".cpu as separate tokens: IDENT, DOT, IDENT, DOT, IDENT instead of scanning it as a single IDENT
• Add Parser.parseSegmentedIdents() to parse segmented (DOT separated) identifiers like "db"."rp".cpu and return them as an []string
• Change AST types that held segmented identifiers in a string to have individual fields. Note that type VarRef still, for now, holds a single string. This may need to be changed.
• Add type HasDefaultDatabase interface to ast.go. Types like type CreateContinuousQueryStatement that have a default database should implement this interface.
• Change parsing functions that work with segmented identifiers to use the new parseSegmentedIdents() function and store segments in the appropriate AST fields.
• Change QuoteIdent function to only quote segments that must be quoted.
• Add IdentNeedsQuotes func that returns true if an identifier would require quotes
• Change Server.ExecuteQuery to check for a default database passed by the caller. If none was provided, see if the statement implements the HasDefaultDB interface and get the default database from there. This fixes a bug in statement normalization.
• Remove uses of influxql.SplitIdent function outside of the parser
• Change Server.expandSources to always return expanded sources in sorted order
• Change ErrDatabaseNotFound and ErrRetentionPolicyNotFound to be functions that report the name of the "not found" object and the error locust

[otoolep]

How about making this a vargs? Then all the callers won't need to be changed.

[otoolep]

Perhaps we should have two versions of this function query and queryRegex. You could factor out the common code, but two functions might be cleaner.

[dgnorton]

Would we also need two versions of queryAndWait then? I wasn't crazy about adding the extra parameter but it seemed like the simplest solution. If we factor into two funcs, every caller has to do the if expected != "" check. No strong preference here...just don't want to add complication.

[otoolep]

Yeah, wasn't sure about this. We can come back it.

[otoolep]

Took a first pass, will re-review again soon. I like the introduction of regex matching into the test suite, very handy. I'm not keen on the new HasDefaultDB interface though, it feels a bit awkward, though I am not in the code so it may be required. I wonder if there is another way, perhaps every statement just have this call, period.

[otoolep]

Minor: document that this is ignored if expected is non-blank.

[otoolep]

OK, I took a look at this. It's obviously a very large change but most of it is adding extra variables to functions for example. The tests haven't changed that much, so the green build gives me confidence too.

[otoolep]

Just to be clear, if we have more than 2 segments here, it's a problem? Normally we can have up to 3 segments. Which segment is not expected to be present?

[dgnorton]

Good catch. That's probably a mistake...not sure why I did that. Will review.

[otoolep]

OK, cool. Might indicate that a unit test is needed to check a specific code path.