fix(platform): bound transform analysis separately from expressions

[yannickmonney]

Transform steps with more than 512 tokens were rejected by the static validator even when their source was modest and valid. Give transform bodies their own bounded analysis capacity: 16,384 UTF-16 units and 4,096 tokens. Expressions and templates retain their existing 8,192-unit/512-token limits; both body and expression depth limits remain 64.

The change uses the existing parser and updates the English, German and French documentation. Synthetic regressions cover exact limits, malformed and hostile inputs, aggregate document limits, and a real CLI release ZIP consumed by the backend validator. No client business code is included.

Validation:

• 99 parser/boundary tests, four CLI release tests and 144 existing flow/type/name/constant tests passed.
• The new CLI regression fails with CODE_SYNTAX when only the old parser is restored.
• Configured scoped TypeScript, lint, format, local SAST, 685 documentation tests and the manual coverage guard passed. All three documentation locales were rendered at desktop and mobile widths.
• Ten synthetic analysis cases were measured in Node 22 and Bun without dispatching authored bodies. Peak observed RSS was about 162 MiB in Node and 138 MiB in Bun; these measurements are not a production concurrency or worst-case guarantee.

Compatibility: the shared validator runs in both the CLI and backend. A newer CLI alone cannot make a release using the larger bodies compatible with an older backend. This PR does not change an existing release tag or deploy a private pack. Full monorepo and deployed acceptance are not claimed by these focused checks.