Various MTU fixes.

[FelixMcFelix]

First, we missed one branch in the overlay action for setting the mtu_unrestricted property on VPC-local packets. This ended up getting missed because the bench/xde-test runner set up the routing table differently to omicron.

Second, when we open up some limited ICMP families for Nexus (intending to support PMTUD), the illumos TCP stack will have its own (and better!) idea about what MSS is actually assigned to a flow. We now reuse this MSS when not explicitly boosting a flow into jumbo frame territory.

Closes #747.
Closes #748.

[FelixMcFelix]

So far, this change has been tested by bringing up a local omicron setup on my helios workstation. Given that this is in single-sled mode, all traffic delivery between guests must go via the guest_loopback pathway. The simplest way to check the new behaviour was to insert a few extra statements into xde_mc_tx_one:

if mtu_unrestricted {
    opte::engine::err!("yay!");
} else {
    opte::engine::err!("nay...");
}

Crude, but we can see this is actually being set now. It wasn't before (but was in cargo kbench):

May  7 16:30:11 farme xde: [ID 130198 kern.warning] WARNING: yay!
May  7 16:33:34 farme last message repeated 2332670 times

Instance connectivity worked fine in this setup. I'm going to pile this up with a few extra OPTE/stlouis changes on a racklette shortly.

[FelixMcFelix]

Some further proof from berlin, using an anti-affinity group to guarantee placement:

BRM42220023 # snoop -c10 -rd cxgbe0 'udp and port 6081'
Using device cxgbe0 (promiscuous mode)
fd00:1122:3344:102::1 -> fd00:1122:3344:101::1 UDP D=6081 S=42966 LEN=8960
fd00:1122:3344:102::1 -> fd00:1122:3344:101::1 UDP D=6081 S=42966 LEN=8960
fd00:1122:3344:102::1 -> fd00:1122:3344:101::1 UDP D=6081 S=42966 LEN=8960
fd00:1122:3344:102::1 -> fd00:1122:3344:101::1 UDP D=6081 S=42966 LEN=8960
fd00:1122:3344:102::1 -> fd00:1122:3344:101::1 UDP D=6081 S=42966 LEN=8960
fd00:1122:3344:102::1 -> fd00:1122:3344:101::1 UDP D=6081 S=42966 LEN=8960
fd00:1122:3344:102::1 -> fd00:1122:3344:101::1 UDP D=6081 S=42966 LEN=8960
fd00:1122:3344:102::1 -> fd00:1122:3344:101::1 UDP D=6081 S=42966 LEN=3096
fd00:1122:3344:102::1 -> fd00:1122:3344:101::1 UDP D=6081 S=42966 LEN=8960
fd00:1122:3344:102::1 -> fd00:1122:3344:101::1 UDP D=6081 S=42966 LEN=8960
10 packets captured

I don't yet have speeds with promisc recorded, but we should get that on dogfood. Results without promisc (via the siphon branch) will be in https://github.com/oxidecomputer/stlouis/issues/744.