Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

Angular's control-directives pipeline treats [formField] as a normal property binding and then inserts separate controlCreate/control instructions. Our Rust pipeline had drifted from that logic and rewrote [formField] into a custom ControlOp carrying the bound value, which emitted legacy output like ɵɵcontrol(ctx.myField, "formField") without ever writing the directive input via ɵɵproperty("formField", ...). In Angular 21 signal forms this leaves FormField.field unset and can surface as NG0950 at runtime.

This change restores the expected control flow for template bindings:
- keep [formField] as a regular PropertyOp
- emit a separate ControlOp after the property update
- reify ControlOp to zero-arg ɵɵcontrol()
- stop extracting duplicate const metadata from ControlOp itself

The tests now cover both the regression and Angular's mixed-order control fixture behavior:
- [formField] must emit ɵɵproperty("formField", ...) plus ɵɵcontrol()
- legacy ɵɵcontrol(value, "formField") output is rejected
- mixed [formField]/[value] bindings preserve update order
- extracted const metadata preserves per-element binding order

Verified with targeted cargo test runs for the new regression, control binding extraction, mixed property ordering, const ordering, pipe slot propagation, and the existing [field] non-control regression.