mgmt-gateway: Update to new MGS serialization of large messages

[jgallagher]

Update payloads and serial console packets are now serialized as raw
data after a hubpack-encoded message, instead of the data being
embedded as part of the hubpack message. Formerly we were required to
allocate several moderately-sized arrays on the stack and copy data
unnecessarily; now the raw data is borrowed from the receive buffer we
pass to net, which requires no extra copies or stack allocation of
large arrays.

This is the hubris half of oxidecomputer/omicron#1645, which did the MGS side of this work.

We get significant improvements in flash and stack, and a modest improvement in ram (although it doesn't look like it at first glance). Before this PR:

# xtask sizes
PROGRAM        REGION  USED   SIZE   LIMIT
mgmt_gateway   flash   26764  32768  32768
               ram     6444   8192   8192

# stackmargin after performing an update, before resetting
ID TASK                STACKBASE  STACKSIZE   MAXDEPTH     MARGIN
11 mgmt_gateway       0x20004000       4096       3416        680

After this PR:

# xtask sizes
PROGRAM        REGION  USED   SIZE   LIMIT
mgmt_gateway   flash   19220  32768  32768
               ram     7456   8192   8192
               
# stackmargin after performing an update, before resetting
ID TASK                STACKBASE  STACKSIZE   MAXDEPTH     MARGIN
11 mgmt_gateway       0x20004000       1536       1048        488

Our absolute ram usage goes up by about 1k, but our buffers are significantly larger: we can now receive ~1k update chunks (previously 512 bytes), and we can send and receive ~1k serial console packets (previously 128 bytes). Handwaving a bit, I think we have about 1.8k more buffer space for only 1k more ram usage.

[mkeeter]

🎉

[mkeeter]

I'm not a huge fan of destructuring in function arguments, but that's just, like, my opinion, man.

[jgallagher]

Hah, me neither. IIRC this is a relic of trying to get cargo fmt to do something reasonable with a long line before I introduced the SerializedMessageBuf typedef; it was doing something heinous like

buffers: &'static mut [ArrayVec<[u8; SOME_LONG_CONSTANT]>;
                       2],

With the typedef I don't think that's a concern any more, so I'll switch this back.