Emit tables for better diagnostics

[dsyme]

The parsers we generate already have some tables for auto-generation of diagnostics

We should extend this to allow better diagnosis of error recovery situations.

Currently for a grammar like this:

nonterm1: 
    | tok1 { }  // ProdIdx 1
    | tok1 nonterm { }  // ProdIdx 2
    | tok2 nonterm { }  // ProdIdx 3
    
nonterm2: 
    | tok1 nonterm { }  // ProdIdx 4
    | tok2 nonterm { }  // ProdIdx 5

The parse will generate states something like this:

state 300:  
    nonterm1: . tok1 
    nonterm1: . tok1 nonterm 

state 301: 
    nonterm1: tok1 .
    nonterm1: tok1 . nonterm  

etc and we emit tables that contain this kind of information:

State --> ProdIdx list ( 300 --> [1; 2], 301 --> [1,2], ....)

ProdIdx --> NonTerm (1 --> nonterm1, 2 --> nonterm1, ...)

For diagnostics we really want textual representations of the states and productions, e.g. tables for:

ProdIdx --> text

   1 --> "tok1 nonterm (line 506 ilpars.fsy)"

State --> text

     301 --> 
         "nonterm1: . tok1 (rule at line 506 ilpars.fsy)
           nonterm1: . tok1 nonterm (rule at line 507 ilpars.fsy)"
     302 --> 
         "nonterm1: tok1 . (rule at line 506 ilpars.fsy)
           nonterm1: tok1 . nonterm (rule at line 507 ilpars.fsy)"

etc.

This information should probably just always be emitted by default - the slight increase in size is of no great concern in modern situations