fix: surface prompt template eval errors, fail-safe async_safety fallback, and set _parent_agent_var in sync wrapper

[Eros483]

What does this PR do?

Three defensive fixes targeting silently swallowed errors and a sync-wrapper
parity gap:

1. agent.py — Agent._resolve_docstring_template() catches eval()
   failures with a bare except Exception: and silently falls back to raw
   {placeholder} text. If a user has a typo in {self.attr}, the broken
   expression ships in the prompt with zero feedback. Added
   logger.debug(..., exc_info=True) inside the except block so template
   errors are visible at debug level.

2. async_safety.py — _is_event_loop_thread() returns True when it
   cannot determine the thread (missing _thread_id attribute). This is
   fail-dangerous: it would block legitimate cross-thread Future.result()
   calls. Changed fallback to False so calls are allowed through when we
   can't verify.

3. method_wrapper.py — create_sync_agent_method_wrapper never calls
   _parent_agent_var.set(self), unlike its async counterpart (line 183).
   Subagents instantiated inside a sync method cannot inherit the parent's
   LLM. Added parent_token = _parent_agent_var.set(self) after the
   call-id push and _parent_agent_var.reset(parent_token) in finally,
   mirroring the async wrapper.

Related issues

N/A

Checklist

• Code follows the project style (uv run ruff check . and uv run ruff format --check . pass)
• Tests added/updated and passing (uv run pytest)
• Docs updated if behavior or public APIs changed
• New source files carry an SPDX license header

Checklist notes

• ruff: Clean on all three changed files.
• pyright: 0 errors, 0 warnings.
• Tests: No new tests required — each change is an internal quality fix
  (debug logging, boolean flip, contextvar parity with existing async path).
  Existing test suite covers all modified code paths.
• Docs: No public API changes. No new files created.

[alessiodevoto]

@dcoapp recheck

[Eros483]

Hi @alessiodevoto . What's up?