refactor(php-transformer): build site plans before report projection

[chubes4]

Summary

Continues #1360 by removing the provisional conversion report from artifact compilation. Internal site-plan construction consumes producer-owned fallback evidence, then the compiler constructs the terminal report once. Public WordPressSitePlan::fromResult() retains strict canonical-envelope validation. Serialized source-report ordering is preserved.

Verification

All ten CI checks passed on 242e71506f4f007754a73cab40462367a1ac6fdb: Composer tests, WordPress integration across PHP 8.2-8.5, and visual parity tools, plus solved-site promotion.

• Added pre-report/final-plan equality coverage for ordinary output and nonempty sanitization fallback evidence, plus rejection of pre-report data through the public API.
• Focused site-plan contract passed on the candidate.
• Initial external-harness comparison of 385 HTML fixtures and nine targeted cases did not catch a final artifact-report regression. CI caught a plan/report asset-count mismatch. The correction retains the completed plan in the final report's transient inputs without changing persisted key order.
• Direct full artifact parity passed all 305 fixtures on baseline and corrected candidate. The original candidate reproduced the CI failure. Added explicit terminal-report counts to the affected commerce/runtime artifact fixture.
• PHP syntax checks and git diff --check passed.
• Local test execution encountered FAIL [empty-plaintext-drops]; separate baseline and candidate runs reproduced the same failure in that environment. Final CI Composer suites passed across PHP 8.2-8.5 as linked above.

Repo-native reproduction, from php-transformer after installing Composer dependencies:

php tests/contract/wordpress-site-plan.php
php tests/unit/text-leaf-element-converter.php
composer test
php tests/parity/run.php

AI Assistance

Implemented and reviewed with AI assistance through OpenCode, coordinated by GPT-6 Astra under Chris Huber's direction. Delegated agents handled implementation and independent review; the coordinator reviewed the diff and required fresh baseline/candidate evidence before publication.