A quick and dirty metric to know if a card can support Vulkan or not - if the card supports OpenGL 4.5 (or OpenGL ES 3.1 if you’re on an SoC) then it’s possible. Doesn’t mean that the vendor will necessarily have a driver for it but someone working on Mesa might for their own reasons.
IR compilers - depending on the Vulkan implementation can take different routes between SPIR-V and the final ISA that’s sent to the card. IIRC on some Mesa implementations it will go from SPIR-V to NIR and then to ISA - it’s been a while so my knowledge on this might be stale. AMD’s open source driver (AMDVLK using LLPC) will go from SPIR-V to GCN directly. It’s not unreasonable to assume that a GPU vendor shares the driver level shader compiler among their API implementations. The driver level shader compiler is generally where one can determine if an architecture is supported or not. If it doesn’t target the architecture’s ISA then no Vulkan.
FWIW: radv’s AOC compiler is quite good, so in the land of hopes and dreams if there’sa driver to target, radv wouldn’t be a bad choice.