gallium: avoid redundant tgsi_exec_machine_bind_shader() calls on draw exec path

gallium: avoid redundant tgsi_exec_machine_bind_shader() calls on draw exec path

tgsi_exec_machine_bind_shader() isn't cheap so avoiding unecessary calls
is a big win.
A similar change should be done for softpipe's fragment exec path but
extra care needs to be taken with the texture sampler state/params.