I'm first trying to get parsing playing nicely without wildcards. 
Whether or not this is a mistake will be seen later.

It is now state table based and tracks input types much more closely, 
greatly reducing code complexity.

Yesterday I realised links (_link) can be referenced before where 
defined but I need to know their column types. So I added a first-pass 
to dig them all out. I can't see a way around that.

With two passes it is likely much could be improved/changed, but for now 
I'll leave the rest of the code structure alone.

Date math gets interesting:

11-Jan-2016-1/10/2016+15.10.2016-4

produces:

DATE_SUB(DATE_ADD('2016-10-15', INTERVAL DATEDIFF('2016-1-11', 
'2016-1-10') DAY), INTERVAL 4 DAY)

Two-digit years <70 ==> 20xx, >=70 ==> 19xx

Unlike pdox 2 + 10-oct-2017 in a date column produces 
DATE_ADD('2017-10-10', INTERVAL '2' DAY) instead of a type error.

Still lots to do, but feels like it is heading in the right direction.

Even as I write this I'm wondering if I shouldn't add lookahead/rollback 
into the stream component so the lexer can give higher-level tokens to 
further simplify parsing .....