wifi: let connect() take a scan result

[mikeysklar]

Stacked on #11308. Base is esp32c5-wifi-5ghz, so only the top commit is new here.

What

CIRCUITPY_WIFI_BAND in settings.toml pins a dual-band radio to one band.

Value	Result
2.4	WIFI_BAND_MODE_2G_ONLY
5	WIFI_BAND_MODE_5G_ONLY
unset or anything else	WIFI_BAND_MODE_AUTO

No new public API. Follows the CIRCUITPY_WIFI_HOSTNAME precedent, read via
settings_get_str() and guarded by CIRCUITPY_SETTINGS_TOML.

Why

AUTO sorts candidate APs by RSSI, and 2.4 GHz usually wins indoors. One room
from my AP, the same SSID connects on channel 5 rather than 149.

Setting	Channel	RSSI
unset (AUTO)	5	-51 dBm
5	149	-57 dBm

So on a same-SSID dual-band network there is otherwise no way to reach 5 GHz.

Design question

@dhalbert suggested a runtime property with an enum value such as BAND_2_4
instead, defaulting to 2.4. That is one IDF call here, but it is public API
needing stubs in the raspberrypi and zephyr-cp wifi ports. Happy to do that
instead if it is preferred.

Hardware tested

ESP32-C5-DevKitC-1-N8R8. Scanned channels per setting:

CIRCUITPY_WIFI_BAND	Channels seen
unset	both bands
2.4	2.4 GHz only
5	36, 44, 48, 149, 157
zzz	both bands

TODO: attach the raw REPL transcript for the table above.

AI assistance

Claude wrote the settings lookup and the docs entry. I ran every band trial
and the connect timings on hardware and checked the channel numbers myself.

[dhalbert]

The reason for a settings.toml value is to specify 2.4 vs 5 when using auto-connect.

I didn't suggest a property necessarily, it could also be an optional arg to connect() or start_ap(), which would limit the way it could change more. I don't know whether Espressif provides simultaneous dual-band connections but we should also look at the Nordic nRF7002, which is dual band.

EDIT: connect(), not start_station(), I think makes more sense.

[tannewt]

Yup, I agree with Dan. This should only impact wifi auto-connect. I think connect can be specific by BSSID or channel already. Channel is especially good because it reduces scan time a ton.

We also considered having connect take a scan result to encapsulate this info. Or maybe ScanResult could have connect() itself.

[mikeysklar]

connect() takes a Network from a scan and uses its bssid and channel.

CIRCUITPY_WIFI_BAND is gone.

As tannewt suggested channel= is much faster (7x). Same for net or bssid=.

ALL_CHANNEL_SCAN is expensive.

10 trials each (connection time measured in seconds):

Call	Mean	Range	Channel
connect(ssid, pw)	11.01	10.77-11.30	5
connect(ssid, pw, channel=149)	1.55	1.14-1.64	149
connect(ssid, pw, bssid=...)	1.62	1.59-1.64	149
connect(net, pw)	1.62	1.59-1.66	149

Scanning to get the Network costs 3.57s, so scan plus connect is 5.18s end to end.