replace nginz disco with native upstream resolver WPB-15302

[jschaul]

WPB-15302

• replace outside upstreams.txt file by an inline block, making use of 'resolve' keyword (available since nginx 1.27.3). Current nix packages pins nginx to 1.28.

nix repl --file nix/default.nix
nix-repl> pkgs.nginx.version
"1.28.0"

(but actually hardcoded nginx to 1.26 for some reason! - So this PR also upgrades nginx from 1.26 to 1.28)

• remove nginz_disco code, and sidecar deployment
• drive-by improvement: be more graceful when shutting down: longer graceperiod, and use quit not sigterm (to process in-flight requests before shutting down - hopefully less 502s that way)

Checklist

• Add a new entry in an appropriate subdirectory of changelog.d
• Read and follow the PR guidelines

[smatting]

where do these settings come from?

[jschaul]

They are kept from before, see https://github.com/wireapp/wire-server/blob/develop/tools/nginz_disco/nginz_disco.sh#L49-L55 which results in current upstream files looking like

upstream cargohold { 
	 least_conn; 
	 keepalive 32; 
	 server 10.100.213.88:8080 max_fails=3 weight=100; 
}
upstream galeb.integrations { 
	 least_conn; 
	 keepalive 32; 
	 server 10.100.18.166:8080 max_fails=3 weight=100; 
}

The only difference here is the addition of a zone, which according to the docs in https://nginx.org/en/docs/http/ngx_http_upstream_module.html (see 'resolve' keyword) is needed, as

The server group must reside in the shared memory.

[pcapriotti]

LGTM