Double-precision floating point, IEEE 754, eight bytes.
A built-in TL type: it has no constructors and is serialised directly. Details are in the protocol description.