docs(insta): custom domains self-heal; read the edge's reason

[tonychang04]

Updates insta/references/deploy.md → Custom domains to match behavior shipping in:

• InsForge/instacloud-compute#398: the plane re-checks custom domains in the background (~every 2 min), and domain check surfaces the edge's actual reason for a stuck hostname.
• InsForge/instacloud-platform#631: with BYO zones, hostnames attached before zone delegate get their records published once the zone is delegated. No re-attach, except a hostname already failed (48h deadline), which keeps the documented one-time domain attach retry.
• InsForge/instacloud-platform#630 (merged): a bought domain reads active once any hostname serves; a failed hostname keeps its own reason.

Adds a short "Troubleshooting a stuck custom domain" list: run domain check, read the error line, map the reason (CAA → 0 issue "letsencrypt.org"; apex not pointed → domain zone delegate / domain delegate or bind www; subdomain not pointed → publish the CNAME DNS-only). Never ask support to "mark it active"; the 60-day ICANN transfer hold never blocks DNS or verification.

CLI surface checked against insta-cli origin/main: domain check prints the plane's errorReason as an error line, domain status prints per-hostname reasons, and domain delegate / domain zone delegate exist. SKILL.md pointer is unchanged (still accurate).

Merge after compute#398 and platform#631 are deployed.

🤖 Generated with Claude Code

---

Summary by cubic

Updates the custom domains documentation to match self-healing behavior shipping in the compute and platform changes.

• The plane re-checks pending hostnames in the background (~every 2 min), and domain check now surfaces the edge's actual reason for a stuck hostname.
• With BYO zones, hostnames attached before zone delegate get their records published once the zone is delegated; only a hostname already failed needs the one-time domain attach retry.
• A bought domain reads active once any hostname serves; failed hostnames keep their own reason lines.
• Adds a short troubleshooting section mapping error reasons (CAA, unpointed apex, unpointed subdomain) to the documented fixes.

Merge order

• Merge after compute#398 and platform#631 are deployed.

^{Written for commit 9c415e1. Summary will update on new commits.}