fix(cloud): clarify plan comparison and raise new service limits

[Hydralerne]

Summary

The deployment capacity dialog could not show four plans in one row. Plan cards also hid metered build usage, and service ceilings came from generic machine presets: Starter allowed only 1 GB per service despite a 6 GB shared pool.

This change widens the comparison, makes build usage and included credits visible, and publishes explicit CPU/RAM ceilings for new subscriptions. The comparison, billing state and engine gates use the same shared resolver.

New offer	Maximum vCPU per service	Maximum RAM per service
Hobby	1	2 GB
Starter	2	3 GB
Pro	4	4 GB
Scale	8	8 GB

One service can use the full shared CPU pool and half its RAM. Pro and Scale workspace CPU ceilings become 4 and 8 vCPU so those advertised sizes can be provisioned. Prices, included credits, total capacity, project/service counts and workload presets are unchanged. Usage remains metered against finite funded credits; these limits do not promise continuous maximum CPU for the entire month.

Contract and UI behavior

• New v4 offers save explicit service ceilings and all capacity fields. Existing paid snapshots keep their purchased terms. Older inherited policies retain their pre-v4 VM CPU ceilings; no subscribers are migrated or upgraded automatically. Unchanged top-ups retain v3 references.
• The existing plan picker and checkout show four paid plans in one row on desktop, two on tablet and one on mobile. Shared capacity, per-service CPU/RAM, build usage, credits, projects and service slots are visible together. The compact welcome summary keeps its project/service/build metrics.
• New offers use shared credits for builds without a separate monthly minute allowance. Older purchased build allowances remain visible and enforced. New copy is translated in all nine locales, and pricing documentation matches the catalog.

Validation

• Full GitHub CI passed on e3678d1a.
• 95 core pricing tests, 357 API billing/resource tests and 77 dashboard billing tests passed. These cover exact CPU/RAM ceilings, oversized requests, native/Compose deployment and runtime gates, saved subscriptions, checkout, reconciliation and SDK/HTTP parity.
• Core, API and dashboard typechecks passed. Public SDK/CLI build and documentation validation passed.
• Dialog layout was inspected in Chromium at 390, 820, 1280 and 1440px, across light/dim/dark, Arabic RTL and German labels, including overflow and keyboard behavior.

Related issue

None.