feat: add NITRIC_HOSTNAME env var for configurable local backend address

[tom-groves]

Presigned storage URLs, gateway addresses, and website URLs are hardcoded to localhost, making them unresolvable when the backend runs in a Docker container addressed from another container.

For example, when calling PreSignUrl from a service in a separate container, the returned URL is:

http://localhost:54321/read/<token>

Which resolves to the calling container itself, not the Nitric backend.

Solution

Introduce a NITRIC_HOSTNAME environment variable (default: localhost) that controls the hostname used in all generated URLs. The value is resolved once at startup in cloud.New() and threaded through storage, gateway, and website service constructors.

Usage:

NITRIC_HOSTNAME=backend-container nitric start

Presigned URL with override:

http://backend-container:54321/read/<token>

Without override (default, no behavior change):

http://localhost:54321/read/<token>

Affected code paths

Location	Before	After
storage.go PreSignUrl	http://localhost:{port}/read/...	http://{NITRIC_HOSTNAME}:{port}/read/...
storage.go PreSignUrl	http://localhost:{port}/write/...	http://{NITRIC_HOSTNAME}:{port}/write/...
gateway.go GetTriggerAddress	[::] → localhost	[::] → {NITRIC_HOSTNAME}
gateway.go GetApiAddresses	[::] → localhost	[::] → {NITRIC_HOSTNAME}
gateway.go GetHttpWorkerAddresses	[::] → localhost	[::] → {NITRIC_HOSTNAME}
gateway.go GetWebsocketAddresses	[::] → localhost	[::] → {NITRIC_HOSTNAME}
websites.go register	http://localhost:{port}/...	http://{NITRIC_HOSTNAME}:{port}/...

Additional hardening

• IPv6 literal hostnames are automatically bracketed for valid URL construction per RFC 3986 (e.g. fe80::1 → [fe80::1])
• Hostnames containing URL-unsafe characters (/?#, newlines, whitespace) are rejected at startup with a clear error
• Added comment clarifying why connectionStringHost (for Postgres) is intentionally separate from NITRIC_HOSTNAME

Note on connectionStringHost

The SQL connectionStringHost in cloud.go is deliberately not changed to use NITRIC_HOSTNAME. It controls how application services reach the Postgres container (a separate container with host-mapped ports), which is a different network path from how callers reach the Nitric backend. The existing RunMode/StartMode logic with host.docker.internal already handles this correctly.