fix embedded bridged responses handling

[dfrycki]

fixes #141

[vmauery]

I helped @dfrycki debug this issue on an OpenBMC platform. When the memmove happens, it was not correctly resetting the offset and shadow offset into that buffer, causing out-of-bounds memory references for encapsulated responses.

[AlexanderAmelkin]

@dfrycki, @vmauery, have you checked it on non-OpenBMC stacks? I don't have any systems with bridging to verify it myself.

[vmauery]

I have tested it on two generations of prior Intel systems, both of which are different IPMI stacks (AMI-based and proprietary). From what we can understand from the behavior of those is that they split the response into two packets and skip this broken path. OpenBMC sends the response all in one packet and hits this issue where the offsets are incorrect.

But now either path works.