Fixes regressions caused by a38e0563.

It never changes.

wined3d/arb: Move fragment program compilation from fragment_prog_arbfp() to arbfp_apply_draw_state().

Rename to *_apply_draw_state() per Henri's suggestion.

Analogous to shader_disable().

We'd like to pass a wined3d_state pointer to fp_enable() to let it handle
actually compiling the fragment pipeline, which doesn't mesh well conceptually
with its use in shader_disable().

Analogous to shader_disable().

We'd like to pass a wined3d_state pointer to vp_enable() to let it handle
actually compiling the vertex pipeline, which doesn't mesh well conceptually
with its use in shader_disable().

Arguably there is a benefit here in that it reduces shader backend state
application down to a single method, although whether that is an improvement may
be up for debate.

More saliently, we would like to move FFP state application out of the state
table and into a linear state function. In practice that function will probably
be vp_enable / fp_enable. In that case we will need those functions, and hence
also shader_select(), to be called when more states are dirty.

This does not in itself mean that shader_load_constants() needs to be merged
with shader_select(), but it does remove the primary impetus for keeping them
separate.

Rename to shader_apply_draw_state() to reflect the adjusted purpose, as
suggested by Henri Verbeet.

Looks like thread is suspended until DBG_CONTINUE is sent on thread
creation debug event.
Signed-off-by: Eric Pouech <epouech@codeweavers.com>

So the allocator matches with documentation.

Setting invalid proxy server has some side effects.

MSVC generates zeroed out entries when EditAndContinue (aka HotPatch)
is enabled.
Signed-off-by: Eric Pouech <epouech@codeweavers.com>