gallium: Add PIPE_CAP_GLSL_TESS_LEVELS_AS_INPUTS

In the Intel backend, it makes the most sense to treat gl_TessLevelInner
and gl_TessLevelOuter as ordinary shader inputs.  For Radeon, it makes
more sense to treat them as system values which get special handling.

We already have a compiler option for this, but the Iris driver will
need a capability bit so we can set it appropriately.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

src/gallium/auxiliary/util/u_screen.c

@@ -263,6 +263,9 @@ u_pipe_screen_get_param_defaults(struct pipe_screen *pscreen,
    case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
       return 1;
 
+   case PIPE_CAP_GLSL_TESS_LEVELS_AS_INPUTS:
+      return 0;
+
    case PIPE_CAP_TGSI_FS_FBFETCH:
    case PIPE_CAP_TGSI_MUL_ZERO_WINS:
    case PIPE_CAP_DOUBLES:

src/gallium/include/pipe/p_defines.h

@@ -807,6 +807,7 @@ enum pipe_cap
    PIPE_CAP_TGSI_CAN_READ_OUTPUTS,
    PIPE_CAP_NATIVE_FENCE_FD,
    PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY,
+   PIPE_CAP_GLSL_TESS_LEVELS_AS_INPUTS,
    PIPE_CAP_TGSI_FS_FBFETCH,
    PIPE_CAP_TGSI_MUL_ZERO_WINS,
    PIPE_CAP_DOUBLES,

src/mesa/state_tracker/st_extensions.c

@@ -334,6 +334,8 @@ void st_init_limits(struct pipe_screen *screen,
 
    c->GLSLOptimizeConservatively =
       screen->get_param(screen, PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY);
+   c->GLSLTessLevelsAsInputs =
+      screen->get_param(screen, PIPE_CAP_GLSL_TESS_LEVELS_AS_INPUTS);
    c->LowerTessLevel = true;
    c->LowerCsDerivedVariables = true;
    c->PrimitiveRestartForPatches =