fix(cli): let plan --restate-model run without local model files - #6219

Description

sqlmesh plan --restate-model fails with doesn't seem to have any models...
on a project that has no local model files, even though sqlmesh run works
there. A restatement runs entirely against environment state (local changes
are deliberately ignored via plan_builder.force_no_diff), so the local
files should not be required.

Reproducer

uv venv && uv pip install sqlmesh && . .venv/bin/activate
sqlmesh init duckdb
sqlmesh plan --auto-apply
rm -rf models seeds tests audits macros logs .cache
sqlmesh plan -r sqlmesh_example.full_model -s '' -e '' --auto-apply
# Error: `('/path/...',)` doesn't seem to have any models...

Creating a dummy stand-in full_model.sql lets the same invocation
succeed and does not overwrite the model definition stored in state,
confirming the restatement really did operate out of state.

Context from @izeigerman on #3260:

Yes, restatements shouldn't require model files now that we purposefully
ignore local changes when doing the restatement.

Changes

1. sqlmesh/cli/main.py — when the invoked subcommand is plan and
   --restate-model/-r appears in the subcommand args, set load = False
   (mirroring how run is handled in SKIP_LOAD_COMMANDS). This
   short-circuits the local model load and the "no models" guard so the
   command can proceed when models/ is empty.

2. sqlmesh/core/context.py — when plan_builder expands
   restate_models, pool environment snapshots with local models before
   matching, so a model known only to state can still be named. This
   mirrors the pattern Selector.select_models already uses
   ({**env_models, **self._models}).

The two together preserve the PlanError("Selector did not return any models") behaviour when the name truly matches nothing in either source,
as shown by the manual test with -r nonexistent.model below.

Test Plan

• Added test_plan_restate_model_without_local_definitions in
  tests/cli/test_cli.py: initialises the example project, backfills prod,
  wipes every local project directory except config.yaml and db.db,
  then runs sqlmesh plan --restate-model sqlmesh_example.full_model --start '' --end '' --auto-apply. The test asserts exit code 0,
  absence of the "doesn't seem to have any models" text, presence of
  "Models selected for restatement", and that model batches were
  executed. Fails on main with the "no models" error; passes with
  this change.
• tests/cli/test_cli.py (the whole file) passes locally (90 passed,
  3 dlt tests skipped — pre-existing optional-dep env issue in my venv).
• tests/core/test_context.py passes locally (110 passed).
• tests/core/integration/test_restatement.py passes locally
  (19 passed, covers the restatement paths through plan_builder).
• tests/core/test_plan.py, tests/core/test_selector_native.py,
  tests/core/test_scheduler.py pass locally (145 passed).
• Manual reproducer confirmed: with the fix, the invocation succeeds,
  sqlmesh_example.full_model is restated to its original rows
  (1, 5), (2, 1), (3, 1), and a bogus -r nonexistent.model still
  errors out cleanly with
  Selector did not return any models. Please check your model selection and try again..
• ruff check and ruff format --check clean on the three touched
  files.

Fixes #3260

Checklist

• I have run make style and fixed any issues
• I have added tests for my changes (if applicable)
• All existing tests pass (make fast-test)
• My commits are signed off (git commit -s) per the DCO

Summary by CodeRabbit

• New Features
  • plan --restate-model can now run when local model definitions are unavailable, using models from the target environment to resolve restatement selections.
• Bug Fixes
  • Restatement plans now include matching models from the target environment alongside locally available models.