(after the thread pool has been improved).

- wrong cut & paste across versions of a given record type
- wrong definition for register_v2

- added preliminary support for VT (virtual table) shape
  type record
- as we don't do C++, it's just a hack to get rid of
  some errors

- create an infrastructure for handling forward usage
  of type records
- used it in some cases (pointer to yet not defined udt,
  function signature, modifiers)

- seperate types' table handling from a single type handling
  by adding codeview_parse_one_type function
- factored all calls for caching symt
- make type parsing helpers return the symt instead of a
  simple error status

- when checking debug info in NB09 or NB11 format (CodeView),
  try to load the global types section

- now correctly parsing parameters types for a function
  signature and storing them in dbghelp internal structures

- no longer caching bitfield information but accessing it directly
  when needed
- removed all kludges for bitfield handling
- reused BOOL parameter to codeview_get_type for monitoring
  traces (will be used in later patches)

- now that we have offset tables to type, we can cleanly
  parse the field-lists and remove the associated kludges
- also adding internal structure for easing parameter sharing
  across functions (codeview_type_parse)
- in mscvpdb.h, also reparated the type records that are
  referenced from other type records (union codeview_type_ref)
  from type records referenced by symbols (union codeview_type)

- enforce the type records that can be accessed by symbols (these are
  anyway the high level ones)

- our engine for parsing types section now requires in entry
  an offsets table for getting directly to each type in the
  section
- (re)construct this table for PDB types

- added definitions for start_search symbol info
- silenced warnings from this symbol

- in MSC debug info parsing, clearly separate a source file
  information from a compiland (including in linetab structure)
- in ELF debug info parsing, now storing compiland directly
  in symtab_elt while browsing the symtab section (we still
  create twice the compilands, once in stabs/dwarf parsing, 
  a second time in symtab parsing)

A void return type is represented by a FC_END/FC_PAD sequence in the
proc format string. Increment the proc offset by two when a function
returns void.

- DBGHELP_NOLIVE env variable is tested earlier to avoid 
  some strange effects in live target detection
- in WineDbg, delete DBGHELP_NOLIVE before attaching to
  a process, because we know it must be a live target