A cycle-accurate simulation implies simulating some particular concrete implementation of the specifications. There are not yet any publicly available implementations of the RISC-V Vector extension [1]. No doubt companies developing such implementations have their own simulators internally, but they are not sharing them (and would be unlikely to after the chips are released).
No doubt there will eventually be Open Source implementations of the V extension, but it might be a while.
It is possible that some implementations might transparently handle a page fault in the middle of a vector load or store instruction, but I would expect most would abort the instruction and set the VSTART CSR to reflect the amount already processed so that the instruction could be re-run and use the freshly loaded page once the page fault has been handled.
You could just buy a Nezha board and experiment with that instead of a simulator. It's $99 and supports Linux and virtual memory.
[1] except the Allwinner D1 chip, available on the "Nezha" evaluation board, which implements an older (0.7.1) draft version of the RISC-V vector spec.