So all files in libFirm do this consistently

It is the job of the frontend to keep loops.
These hacks here only worked for some, but not all cases.
In particular a more clever SSA-construction algorithm might not place redundant PhiMs, which would render memory before memory-less loops unkept.

get_irn_op_() checks the same.

This means floatingpoint calculation in a firm graph must be performed
in the mode specified (and may not be performed in a wider mode as
before).

Still broken for anything except int f(void) { return 0; } though...

Get away from the strange set_default_XXX_op style and instead let the
code modules register their node ops directly.

The opcode list was a member of irprog before which wasn't really
handled consistently. Also make sure opcodes are properly freed at
ir_finish().

Now every bit in the public API is documented so we can enable doxygen
warnings for undocumented members.

- Put things into logical groups
- Add cross references from node overview to API docu

memops are nodes that have memory inputs, you can generically query them
for their memory input. We can also get rid of get_fragile_op_mem in
favor of get_memop_mem.

We simply always assume that Sel can only return a null pointer if
its input is a null pointer.

This should fix all known bugs, make the code easier to read and more
robust.

This is the new way of handling switch-jumps. The node contains a table
which maps (ranges of) input values to proj numbers. Compared to a
Cond-node this results in a clean consecutive sequence of Proj numbers
(no searching for a free number for the default_pn anymore) and allows
factoring multiple cases jumping to the same block in a single Proj
(though we still need the optimisation in cfopt for that).