fix(a2a): percent-encode context id fields so ids with separators round-trip

[anxkhn]

Please ensure you have read the contribution guide before creating a pull request.

Link to Issue or Description of Change

1. Link to an existing issue (if applicable):

• N/A (no dedicated issue; the bug is described below per the issue-template structure, as CONTRIBUTING allows for non-doc PRs).

2. Or, if no issue exists, describe the change:

Problem:

The A2A context id codec in src/google/adk/a2a/converters/utils.py is not
invertible when app_name, user_id, or session_id contains the /
separator. _to_a2a_context_id joins [ADK, app_name, user_id, session_id] on
ADK_CONTEXT_ID_SEPARATOR ("/"), and _from_a2a_context_id splits on "/"
and requires exactly four parts, otherwise it returns (None, None, None). So
when any field contains /, the join produces more than four segments and the
inverse silently fails:

from google.adk.a2a.converters.utils import (
    _to_a2a_context_id,
    _from_a2a_context_id,
)

cid = _to_a2a_context_id("app", "user", "projects/p/sessions/s")
print(_from_a2a_context_id(cid))
# expected: ('app', 'user', 'projects/p/sessions/s')
# actual:   (None, None, None)

A returned (None, None, None) means A2A session resolution silently fails for
those ids. Fully-qualified resource paths are a realistic source of embedded
/ (for example resource paths used as a user_id, or an Agent Engine session
path used as a session_id).

The decode path also had a stale docstring that still described the old
ADK$app_name$user_id$session_id format, and an unreachable except ValueError
around str.split (which never raises), which masked the failure mode.

Solution:

Percent-encode each field on encode (quote(field, safe="")) and decode it on
decode (unquote(...)). This is the smallest change that:

• keeps the exact four-part ADK/<app>/<user>/<session> structure (so the
  existing "too many parts" rejection still holds),
• is invertible for a separator in ANY field (a bounded split(sep, 3) would
  only fix session_id, not user_id/app_name, and would break the existing
  too-many-parts test),
• leaves ids without special characters byte-identical to the previous format,
  so there is no breaking change on the wire for the common case.

I also fixed the stale $-format docstring and removed the dead
except ValueError.

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

Added a parametrized regression test
test_roundtrip_context_id_with_separator_in_ids covering / in session_id,
in user_id, in app_name, and a multi-slash session id. It fails on the
current code (each case returns (None, None, None)) and passes with the fix.
Also updated test_to_a2a_context_id_special_characters to expect the
percent-encoded form (%40 for @).

$ pytest tests/unittests/a2a/converters/test_utils.py -q
24 passed

$ pytest tests/unittests/a2a/ -q
366 passed

pre-commit run --files src/google/adk/a2a/converters/utils.py tests/unittests/a2a/converters/test_utils.py
passes (ruff, isort, pyink, addlicense, ADK compliance checks) with no
reformatting.

Manual End-to-End (E2E) Tests:

This is a self-contained, pure-function correctness fix in the A2A context id
codec with no runtime/UI surface, so it is exercised by the unit tests above
rather than an adk web/runner flow. The snippet in the Problem section is a
minimal manual repro (returns (None, None, None) before the fix, the original
ids after).

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have added tests that prove my fix is effective or that my feature works.
• New and existing unit tests pass locally with my changes.
• I have manually tested my changes end-to-end.
• Any dependent changes have been merged and published in downstream modules. (N/A)

Additional context

The percent-encode approach was chosen over a bounded left split because it fixes
every field (not just session_id) and preserves the existing four-part
validation (including the "too many parts" rejection test). Ids without special
characters are unchanged on the wire.