prove single-component output is byte-identical to the pre-component baseline

[joshua-temple]

Problem / why

The foundation's keystone safety property is that adding the component model changes nothing
observable for a single-component repo. Every existing single-component repo must stay green
with byte-identical generated workflows and byte-identical state serialization, because its
manifest has no components: block and resolves to the single implicit default component. A
multi-component example must load into the component-major model and serialize back
round-trip-clean. This is the v1-freeze gate test before the multi-component surface is built
on top.

Scope

This PR does:

• Run the existing fleet and e2e suites against the component-major spine and confirm green
  across every current single-component example repo, with empty generated-workflow diffs and
  empty serialized-state diffs.
• Add a multi-component round-trip e2e: a minimal manifest that declares components: loads,
  generates per-component files, and serializes its state back to
  state.components.<name>.<env> round-trip-clean.

This PR does NOT:

• Add a full multi-component capability example (that is V4); the example here is a minimal
  round-trip fixture.
• Add any new capability; it is purely the foundation proof gate.

Files / areas touched

• e2e/ scenarios (single-component stability, multi-component round-trip).
• A minimal multi-component fixture manifest.

Tests and coverage required

• Empty-diff assertion on single-component generated workflows and serialized state.
• Multi-component round-trip e2e with a stable-shape assertion.
• Full existing fleet green.

Acceptance criteria

• Every existing single-component fleet repo passes unchanged with empty diffs (zero manifest
  edits).
• A multi-component example loads, generates, and serializes round-trip-clean.

Dependencies

Blocked by F3 and F4 (and F2 for the multi-component fixture).

[joshua-temple]

Implemented and merged via #515.