[wushunhao]

Summary

Port of the error-message part of project-openan/a2a-t-sdk-python#59. The built-in fallback for missing local_file business content already exists in the Java SDK, so this PR only carries the follow-up message fix.

When a template is missing, template.not_found used to claim the language was unsupported even when the templateUri itself did not exist, misdirecting callers.

• Reword the bilingual message to a plain does-not-exist:
  • zh-CN: 模板「{template_uri}」不存在
  • en-US: Template '{template_uri}' does not exist
• The declared fact parameters (template_uri, language) and the error-code contract are unchanged; the language is fixed configuration, so the message no longer echoes it while the facts still carry it for programmatic diagnosis.
• Clarify the ErrorCatalog Javadoc that a message template may render a subset of the declared facts.
• Sync the zh/en API reference error tables.
• Add ErrorMessagesResourceTest pinning the rendered message in both languages; the retained facts are pinned by LlmBackedPromptMetadataExtractorTest.

Verification

• ErrorMessagesResourceTest: 2 passed
• LlmBackedPromptMetadataExtractorTest: 9 passed
• mvn -B clean verify -pl '!a2a-t-corpus' (the CI command) on a clean checkout of the head: BUILD SUCCESS for every module
• Reviewed by three concurrent review passes (correctness, tests/CI, docs/port fidelity) with no blocker/major findings

Signed-off-by: w00609601 wushunhao1@huawei.com

[wushunhao]

Review and follow-up

Three independent review passes were run in parallel with distinct lenses (correctness/contract integrity, tests/CI, docs/port fidelity). All three returned APPROVE with no blocker or major findings. Two minor findings were raised and have been addressed in 82fa5cdd:

1. ErrorCatalog.TEMPLATE_NOT_FOUND constant Javadoc still framed the failure as a language problem. Reworded to describe the trigger (the template is missing for the configured language) and what the message renders (the template URI only).
2. ErrorMessagesResourceTest carried a tautological third case with an over-described Javadoc. It was dropped: the retained language fact is already pinned at runtime by LlmBackedPromptMetadataExtractorTest.

Verification

• ErrorMessagesResourceTest: 2 passed
• LlmBackedPromptMetadataExtractorTest: 9 passed
• mvn -B clean verify -pl '!a2a-t-corpus' (the exact CI command) on a clean checkout of 82fa5cdd: BUILD SUCCESS for every module
• CI checks: Build & Test, DCO, EasyCLA all pass

Notes

• Port fidelity: the message wording matches a2a-t-sdk-python#59 (b4f9d795) exactly, and the declared fact parameters (template_uri, language) plus the error-code contract are unchanged. Java has no corpus mapping file on main and no placeholder-equality gate to relax, so those parts of the Python change have no Java counterpart.
• Design docs: no spec/impl update was needed — this is a bug fix that does not change design facts (code, trigger, aliases all unchanged).
• Known nit: the first commit message still says "pinning ... the retained language fact" (the phrase was written before the third case was dropped). The final state is covered by the later commits; history is left as-is rather than force-pushed.