docs: trim the configuration page

[7174Andy]

Stacked on #494 — review that one first; this PR targets its branch, so the diff here is the trim alone.

Summary

1123 → 758 lines. Same treatment as the DNS page in #492: cut what was duplicated, cut what was false, leave the reference intact.

The page compared the same five scenarios in three separate tables, and documented three mechanisms that don't exist.

Removed as duplicated

• Full Configuration Examples (303 lines) was five complete configs in which 8 of 11 top-level sections were byte-identical. Only dns, eip and ssl varied — which is exactly what the "Key Differences Between Examples" table below them tabulated, and what "Choosing a Configuration" above them tabulated again. Now one base config plus a per-scenario overlay, with the three tables collapsed into one.
• The four per-provider YAML snippets under SSL Providers, restating those same scenarios a fourth time.
• "Let's Encrypt Rate Limits" — a second copy of the warning admonition 40 lines above it. Its two extra numbers moved into that admonition.
• "Browser Access" — restated troubleshooting.md#reaching-the-allocator, which it already linked to.
• "Use Your Own Research Software" — adapting.md owns this.
• The 70-line schema dump under Allocator Configuration. Every key and default is in the reference tables immediately below it; replaced with a section skeleton.

Removed as false

Section	Claim	Reality
Configuration System	Env-var overrides via ENV_VAR, and command-line overrides	No @hydra.main, no argv handling; get_config() calls compose() with no overrides. Neither exists.
Environment Variables Not Working	Debug with echo $DB_PASSWORD	Nothing reads DB_PASSWORD from the environment — it appears only inside a log string
Multiple GPU Types	python main.py --config-name=config-gpu	Same reason: not a Hydra app, so this does nothing

Configuration System now says what is true: edit the file, and CONFIG_DIR / CONFIG_NAME move which file is read.

Also dropped "Configuration Best Practices" (five generic bullets) and the two remaining Troubleshooting Configuration entries, which troubleshooting.md covers.

Untouched

Every reference table other pages anchor into — db, machine, dns, ssl, monitoring, manual, machine-type-options — plus the Cloudflare Tunnel walkthrough, AMI ID, and Deployment Identity.

Testing

• The examples actually work. Composed the base with each of the four overlays and ran both validators on the results:

  Scenario	lablink-validate-config	CLI validate_config
  IP Only	[PASS]	OK
  Let's Encrypt	[PASS]	OK
  CloudFlare	[PASS]	OK
  ACM + ALB	[PASS]	OK

• Anchors: parsed every built page in site/ for links into configuration/#… and confirmed each resolves against the rebuilt page's heading ids — no missing anchors anywhere in the site. (Caught one on the way: ### ACM + ALB slugifies to acm-alb, not acm--alb.)

• YAML: all remaining blocks validate against the Config dataclass; down from 19 blocks to 12.

• Build: mkdocs build clean, same pre-existing griffe warnings as main.

Design Decisions

• Base + overlay rather than deleting the examples. An operator wants a file to copy. Splitting into base + a dns/eip/ssl overlay keeps that while removing the eight sections that were being maintained in five places — and the overlay is the comparison, so the diff table stops needing to exist separately.
• Kept the section skeleton at the top. A reader landing on "Configuration Files" wants the file's shape immediately; a 17-line skeleton gives that without duplicating the reference tables underneath.
• Left the reference tables completely alone. They're the page's job and eight other pages link into them. The trim is entirely prose, examples, and dead sections.