Declare distributed outputs and clarify scenario authoring

[VEZY]

Cross-object outputs now belong to the model's outputs_ declaration, using Distributed(Default(...)) or Distributed(Required(...)). Scenarios choose destinations with anonymous OutputTo entries, and kernels request destination columns by variable name. This keeps a model's output definitions separate from the choice of plants or organs that receive them.

The compiler validates destination declarations, required initial values, and output ownership. A single destination can infer its variables; multiple destinations must explicitly partition them. The graph editor, generated scenarios, tests, benchmarks, examples, and authoring skill use the same API.

This PR also:

• Separates stable object IDs from optional display names. Selectors use IDs, while the graph editor shows the ID when no display name is supplied.
• Adds weather datetimes to collected outputs and a compact default display for Authoring.validate_scenario, while retaining structured diagnostic details.
• Clarifies the documentation on coupling, model switching, hard calls, metadata, and PlantGeom visualization, and explains how a common model core can support crop and functional-structural approaches.
• Integrates the status-preparation optimizations from [codex] Avoid rebuilding unchanged statuses during compilation #215 and updates their regression fixtures to the distributed-output API. Includes the workflow updates from build(deps): bump actions/upload-artifact from 4 to 7 #217 and build(deps): bump actions/checkout from 6 to 7 #218.

This is an intentional API change: distributed output defaults and requirements move out of scenario destination declarations and into model schemas. Existing scenarios and kernels must use the new OutputTo and output_targets forms documented here.

Validation

The full local Julia package suite passed on 520d903: 5,579 tests, zero failures or errors, using an unfiltered Kaimon-managed package test run. Frontend type checking and all 25 frontend tests passed. The full local documentation build and static-export validation passed for 58 pages, together with three documentation-renderer regression tests.

On GitHub, the graph editor browser tests, documentation build, and all three downstream integrations (XPalm, PlantGeom, and PlantBiophysics) passed on 520d90387c4042c541ef122e92b0e73b553f01d4. The XPalm pass confirms that this API resolves the mismatch seen when current XPalm is tested against the older API in #215.

Julia 1.10 CI exposed four allocation-test failures: the measurement included 64/160 bytes of boxing at the test's dynamic call site. A local reproduction confirmed that the assignment itself allocates zero bytes. Commit 1f6cda76 moves the measurement into a function with concrete argument types, keeping all zero-byte assertions and value checks unchanged. The complete output-target API test file then passed 215/215 on both Julia 1.10.11 and 1.12.1. No runtime code changed for this fix; the full package suite was not repeated locally after this test-only correction.

The final merge from main changes merge history only: its tracked files are identical to 1f6cda76. GitHub CI has restarted for final head 2a1084fe058db0c62bbece6e01e6d5b56add0b52; those results are separate from the completed checks above.

The benchmark workflows still fail during dependency resolution before taking measurements. They do not provide a performance result for this PR.