PHP use-import map 2/5: keep written qualified receiver types (stop flattening FQNs)

[filipechagas]

Parent

Spec: #18 (ticket 2/5).

What to build

Stop flattening written PHP type annotations to short names. Today _php_name_text (engine.py:671-675) does rsplit("\\", 1)[-1], so private \Vendor\Sdk\Client $c; is indistinguishable from private Client $c;. Only inline-new keeps the written form (receiver_qualified, engine.py:4903).

Thread a qualified form (e.g. receiver_type_qualified) alongside the short name through the receiver-type table for properties, constructor-promoted params, ordinary params, and locals (engine.py:671-675, :717-740, :761-885 _php_method_receiver_types).

Constraints

• Widest-blast-radius ticket of the five: it changes the receiver-type table's value shape. Every consumer of that table must be audited (the PHP: union- and intersection-typed receivers still mint a same-file bare-name edge (user story 11) #9 fix may have touched the same sites — rebase on latest v8 first).
• Short-name behavior must be bit-for-bit unchanged for unqualified annotations; the qualified field is additive.
• On its own this ticket already lets a fully-qualified annotation resolve or refuse correctly — cover that with a test if reachable without PHP use-import map 3/5: PhpNameResolver decisive refusal (closes #16) #21, else assert the stamped fact shape.

Acceptance criteria

• private \Vendor\Sdk\Client $c; stamps both the short name and the written qualified name on the raw-call fact
• Same for constructor-promoted params, ordinary params, and local variable annotations
• Unqualified annotations produce facts identical to today plus (at most) an absent/None qualified field
• All existing PHP member-call tests pass unchanged
• Full suite green vs the current v8 baseline (reconcile counts; baseline at 102ae79 was 4024 passed / 36 skipped)

Blocked by

#19 (shared parser/metadata groundwork) — soft dependency; can be developed in parallel but must merge after it.

[filipechagas]

Closed by PR #31. Orchestrator gate evidence: independent full-suite runs reproduced 4068/36 at branch head (pre-#26 baseline 4057/36 + 11 new tests) and 4072/36 on the merged tree (post-#26 baseline 4061/36 + 11). Parity additionally evidenced by a byte-identical extract() graph against v8 over a corpus covering all four annotation positions, and a tree-wide 782-blob merge verification. v8 baseline is now 4072/36. #21 is fully unblocked (deps #19, #20, #26 all merged).