spirv-val invalidated a fuzzed shader because a 64-bit float type was declared but the Float64 capability was missing.

error: line 67: Using a 64-bit floating point type requires the Float64 capability.
  %double = OpTypeFloat 64

Debugging spirv-fuzz, I found that TransformationAddTypeFloat adds 16-bit and 64-bit float types without adding the corresponding Float16 and Float64 capabilities.