DHCP refactor

[Dirbaio]

• DHCP is now a Socket, so it can process and emit packets on the fly instead of buffering them. Reduces RAM requirements by 1.5kb. Fixes Convert DHCP client to a socket #446, DHCP replies with ICMP Port Unreachable when RawSocket buffer full #457
• No longer requires access to the Interface, it stores all state internally. Poll now has no params.
• Now handles DHCP NAK packets
• Now handles ACK packets only in states where we expect them.
• Renew retransmission algorithm correctly implements RFC2131: waits for (expires_at-now)/2 before next retry.
• Retry REQUESTs slower and with exponential backoff. Fixes DHCP retransmits request too fast, breaks if server is slow. #464
• Remove 0.0.0.0s from the DNS serevr list. (TP-link routers seem to pad the DNS server list with 0.0.0.0 to a fixed size.. wtf?)

Tested against these DHCP servers:

• mikrotik hAP AC2, RouterOS v6.45.9 (long-term)
• dnsmasq 2.84rc2 (linux router)
• Askey rtf8115vw (Movistar fiber, Spain)
• TP-Link TD-W8970
• Linksys E2000 with Tomato 1.28
• Pixel 4a Android 11 USB-Ethernet tethering
• Ubiquiti EdgeRouter X, many thanks to @adamgreig
• TP-Link Archer C7 v2 on OpenWrt 19.07.3, many thanks to @GrantM11235

Testing on more routers would be highly appreciated, especially crappy ISP routers. It requires a linux machine with wired Ethernet connection, and only 5 minutes of your time! Instructions here

[Dirbaio]

Marking as draft until I remove the "fast renew" hack commit

[ryan-summers]

This is all looking much nicer - thanks for putting in the work to get this refactored.

I'm going to test integrating this into our hardware platform and verify that things function as intended shortly, and will let you know how things go. Other feedback is attached in-line below - most is general questions on design decisions. I don't believe I found anything problematic during review.

[ryan-summers]

Follow-up, I have now tested this on Stabilizer and monitored network traffic with Wireshark. I can confirm that the DHCP client renewed after 4 minutes (on an 8 minute lease) and the ICMP message is absent as well.

I also want to note that the buffer removal is also quite significant and this refactor makes DHCP much easier to use. Thanks for the hard work here!

[ryan-summers]

@Dirbaio Ping on this - is there any chance we can work towards getting this merged? This is a very nice improvement to DHCP.

[Dirbaio]

@ryan-summers Thank you for the review! and sorry for the delay, I've had a really bad month.

Addressed all the comments, could you take another look?

[ryan-summers]

No worries! I appreciate the great work on the refactor. I believe there's only one discussion still in place (re: DHCP source/dest ports), but other than that, the changes look good

[Dirbaio]

Changed the port check to a hard assert, I believe this should be now good to go.

... except the CI now fails due to a defmt semver-breaking change :( knurling-rs/defmt#494