reverseproxy: Add ID field for upstreams

[francislavoie]

First step for #4341, Caddyfile support would be needed to complete it. /cc @ulrichSchreiner

This makes it possible to specify upstreams in JSON with an "id" field, to disambiguate upstreams with the same dial address:

"upstreams": [
    {
        "id": "second",
        "dial": "httpbin.org:443"
    }
],

Given the JSON config from https://caddy.community/t/healthcheck-not-working-with-multiple-servers/13563, with just "id" fields added, we see that it works, the internal state has separate upstreams:

$ curl -s localhost:2019/reverse_proxy/upstreams | jq
[
  {
    "id": "second",
    "address": "second",
    "healthy": true,
    "num_requests": 0,
    "fails": 0
  },
  {
    "id": "first",
    "address": "first",
    "healthy": false,
    "num_requests": 0,
    "fails": 0
  }
]

To avoid it being a breaking change, I left in the "address" field in the admin API, deprecated for now, to be removed at a later date. It's more accurate to call it "ID" even though it could be an ID, a dial address, or a SRV address, depending on how it was configured (i.e. whatever the upstream's String() function returns).

[francislavoie]

I made a second commit with a proposed way to specify IDs in the Caddyfile, by using | as a separator. I consider this part still up for debate though. I think it's an okay solution, but 🤷‍♂️

[mholt]

Ok, so I appreciate the thought going into this. I think I want to see about coming up with a different fix though that doesn't require the user to name the upstream separate from the upstream's... well, its name (address, like DNS name or IP). I will try to give it some brain cycles this week.

[mholt]

(With all due respect) I will be sunsetting this PR in favor of #4470 which will properly distinguish health check state from upstreams, so IDs will not be needed. (We can reopen if that's not the case for some reason, but I'm pretty sure this will be superseded by that.)

Thank you for working on it. 🙏

[francislavoie]

@mholt I don't see how #4470 will distinguish the upstreams from two different reverse_proxy handler instances. There's still no identifier to disambiguate two hosts with the same upstream address in var hosts = caddy.NewUsagePool(). That's the crux of this issue. upstream.fillHost() will still return the same host for both, so they'll still share the health state. The only way they'll return something different is if upstream.String() returns something different for the two instances. This is what this PR does, it allows for an optional ID to be specified to disambiguate them.

[mholt]

I'm not finished yet. When I am, the hosts in the pool will not be storing their health state. The crux of the problem will be fixed. Edit: Done now.