1) There exists no pseudopotential generated with the TB09 MGGA exchange-correlation functional, because it is not possible to compute the XC total-energy with that functional. Consequently, the MGGA cannot be used to generate a pseudopotential, and one must therefore use a pseudopotential generated with a different functional.
2) We have chosen to use LDA pseudopotentials for MGGA calculations by default, but you can in principle also use GGA potentials. I'm not sure how much of a difference it will make.
3) For geometry optimization prior to a MGGA calculation, we usually go for a GGA functional, e.g. PBE or PBEsol. The PBEsol (denoted "GGA.PBES" in ATK) will often give rather accurate lattice constants as compared to experiments.