InterMix 10: split dev and compiled production DI runtimes

[abmmhasan]

InterMix 10 performance architecture and implementation branch.

Primary goal: maximize sustained successful request throughput while preserving DI functionality, following PHPForge engineering principles.

Architecture direction:

• explicit development path: optimized dynamic/reflection runtime; no compiler required
• explicit production path: finalized/compiled generated PHP runtime
• same observable DI semantics across development/production boundaries
• immutable build-time definition graph and compiler planning
• direct compiled dependency edges rather than recursive public string-based get() calls
• specialized singleton/transient/scoped paths and scope seeds
• targeted runtime islands only for genuinely runtime-dependent behavior
• compiled tags and lifecycle/scope-leave specialization
• production deoptimization with singleton/scoped identity transfer
• deployment-time artifact validation and prevalidated loading
• zero Repository/Reflection/resolver machinery in fully static generated artifacts
• CacheLayer remains optional
• Closure serialization remains explicit and is never implicit in normal DI resolution

Architecture draft: docs/intermix-10-performance-architecture.md

Authoritative final completion checklist: docs/intermix-10-final-checklist.md

InterMix standalone status: COMPLETE.

• all 33 InterMix architecture checklist items resolved
• PHP 8.4/8.5 stable + lowest QA matrices green
• PHPStan/Psalm green on PHP 8.4 and 8.5
• clean production install green
• native PHPForge benchmark jobs green on PHP 8.4 and 8.5
• no project-level ic:* command override
• no temporary diagnostic workflow
• final parity/error-surface and fully-static-artifact tests included

Representative final benchmark snapshot:

• PHP 8.4: native transient 0.203 µs; generated static transient 0.367 µs; dynamic transient 20.031 µs; static warm singleton 0.117 µs
• PHP 8.5: native transient 0.194 µs; generated static transient 0.356 µs; dynamic transient 20.112 µs; static warm singleton 0.118 µs

The immutable ID→slot candidate was benchmarked but not adopted: its nominal gain was not stable enough across the non-stable CI runners to justify additional production representation complexity. The benchmark remains as evidence.