- Fix string support.
 This would make strings a special kind of arrays flagged by the 
 [string] attribute. Pidl itself would support a couple of extra 
 attributes for it's own use while being compatible with other IDL 
 compilers. 
 Proposed extensions for pidl (to arrays):
   [convert(t)] attribute for forcing conversions from CH_UCS2, etc to UTF8
   [noheader] attribute -> Indicating the string is not preceded

Implement:
  ndr.pm:
   - represent a string as an array with length set to "STRING"
   - [string] implies is_varying unless noheader is specified.
  ndr_parser.pm:
   - if [charset()] specified, use instead of the for loops (call ndr_pu{ll,sh}_charset()
   - calculate length using helper function if [string] is specified

The various flags for strings would change as follows:
 
LIBNDR_FLAG_STR_ASCII               -> [convert(CH_ASCII)]
LIBNDR_FLAG_STR_LEN4			    -> [string]
LIBNDR_FLAG_STR_SIZE4	            -> [size_is()] or if needed [conformant]
LIBNDR_FLAG_STR_NOTERM	            -> array
LIBNDR_FLAG_STR_NULLTERM            -> [noheader]
LIBNDR_FLAG_STR_SIZE2	            -> uint16 length; [string] char data[length]
LIBNDR_FLAG_STR_BYTESIZE            -> uint16 length; [string] char data[length]
LIBNDR_FLAG_STR_FIXLEN32            -> [32]
LIBNDR_FLAG_STR_CONFORMANT          -> no longer needed
LIBNDR_FLAG_STR_CHARLEN	            -> ???
LIBNDR_FLAG_STR_UTF8                -> Nothing (but UCS2 has [convert(CH_UCS2)]
LIBNDR_FLAG_STR_FIXLEN15            -> [15]

- True multiple dimension array / strings in arrays support (closely related to 
	things specified above)

- compatibility mode for generating MIDL-readable data:
 - strip out pidl-specific properties
 - convert subcontext() to an array of chars.