fix: isolate retry hook callback diagnostics

[WSxDemise]

[CodePuppy Agent] - ## What

Add a synchronous streaming_retry_display_suffix callback seam to the streaming retry warning path.

Why

Retry warnings can benefit from deployment-specific diagnostics without putting provider or enterprise details into the core harness.

How

• Add a provider-neutral retry-warning suffix callback phase.
• Preserve generic warning output when no callback is registered.
• Collect additive suffixes in registration order.
• Isolate callback exceptions, invalid returns, and oversized values so plugins cannot interrupt retry recovery.
• Add walk_exception_tree() for shared traversal of wrapped causes, contexts, and ExceptionGroup members.
• Support callable objects safely in callback diagnostics rather than assuming callbacks always expose __name__.

Testing

• /pr-me local gate: changed-file Ruff and formatter checks passed.
• /pr-me affected tests: 209 passed.
• /pr-me independent pre-PR review: clean.
• Full validation previously passed: 12,692 passed, 81 skipped, 1 xpassed.
• i18n audit completed.

Scope

Provider-neutral OSS infrastructure only. This PR contains no Walmart, Element, gateway response-body, or private diagnostic behavior.