fix: make unwind2 more resilient

[JonasKruckenberg]

This PR makes the stack walking implementation of unwind2 more resilient and useful outside of panics. The big fix here is that apparently leaf functions do not spill their return address to the stack instead keeping it in register and EH_FRAME has no explicit instructions on how to recover that. At least according to LLVMs libunwind
https://github.com/llvm/llvm-project/blob/6de5d1e46d1812de2bbbbe8d8d2c811e4d16acbe/libunwind/src/DwarfInstructions.hpp#L295-L296
Kind of annoyed that this was so obscure and difficult to find :/

In the way we previously used stack walking this was fine since save_context always explicitly pushed the return address onto the stack:

k23/libs/unwind2/src/arch/riscv64.rs

Lines 216 to 219 in 4c7edf0

	add sp, sp, -0x210
	.cfi_def_cfa_offset 0x210
	sd ra, 0x200(sp)
	.cfi_offset ra, -16

In essence when called through begin_unwind there never was a leaf function that would show this behavior.

But now with this fixed we can finally construct stack traces from inside the kernels teap handler too! (In the trap handler we don't control where the return address lives, so need to deal with it potentially not being on stack at all).