fix(deps): require pydantic-ai 2.27.1 so tracing excludes retry content

[Ethan-Xingyue]

Which issue or RFC does this PR close?

Closes #1228.

Rationale for this change

include_content=False did not cover the schema-validation retry path: the retry feedback quoted the
model's raw output into gen_ai.input.messages and pydantic_ai.all_messages, and for Memory extraction
that output is the proposed Memory content. Upstream fixed it in 2.27.1, but our floor was >=2.14.1, so
a fresh install could still resolve a leaking version.

What changes are included in this PR?

• Raise the pydantic-ai-slim floor to >=2.27.1 and relock (resolves to 2.29.0).
• Drop the now-obsolete retry-exception paragraph from the Phoenix how-to, docs/en and docs/zh.

Are there any user-facing changes?

The minimum pydantic-ai-slim is now 2.27.1, and the tracing guarantee from #1227 holds without
exception. No API or persisted-format changes.

How was this change tested?

make test, make docs-test, and make check pass. Leak checked with an in-memory span exporter
through the real extraction path: 2.14.1 leaks those two attributes, 2.29.0 leaks nothing, both with the
retry confirmed triggered.

[Ethan-Xingyue]

Added a regression test so this can't come back: the model first returns output missing a nested required field, and we assert that text never lands in a span. The output type has to be nested — a flat one gives a loc of length 1, which upstream already excluded. Passes on 2.29.0, fails on 2.14.1.

[PsiACE]

LGTM, thanks