Make RedisException the canonical error contract for all public Redis operations

[willronchetti]

Problem

Redis driver exceptions leak out of dcicutils' public Redis API. Every RedisBase method,
create_redis_client, and every RedisSessionToken operation except store_session_token
propagated redis.exceptions.* straight to the caller, so a consumer that wanted to survive an
unreachable or failing Redis had to import redis-driver exception types itself.

Verified against the pre-change code with an injected redis.exceptions.ConnectionError:

RedisBase.get           -> LEAKS redis.exceptions.ConnectionError
validate_session_token  -> LEAKS redis.exceptions.ConnectionError
delete_session_token    -> LEAKS redis.exceptions.ConnectionError
from_redis              -> LEAKS redis.exceptions.ConnectionError

Change

dcicutils.redis_utils gains a public translate_redis_exceptions decorator (and the
REDIS_DRIVER_EXCEPTIONS tuple it uses). It is applied to:

• create_redis_client
• every public RedisBase method: info, set, get, set_expiration, ttl, delete,
  hget, hgetall, hset, hset_multiple, dbsize
• every RedisSessionToken operation: from_redis, store_session_token,
  validate_session_token, update_session_token, delete_session_token

Driver failures — redis.exceptions.RedisError and redis.exceptions.RedisClusterException
(which deliberately sits outside the RedisError tree) and their subclasses — now raise
RedisException, with the original attached as __cause__ for callers that want to inspect it.

Programmer errors (TypeError, AttributeError, ValueError, KeyError) are deliberately
not translated, so real defects are not masked. Nesting decorated calls is idempotent: an
already-canonical RedisException passes through unwrapped.

Snovault should consume this public error contract rather than importing Redis-driver
exceptions. RedisException is importable from both dcicutils.redis_utils and
dcicutils.redis_tools; catching it is sufficient for every operation listed above, and
Snovault's configurable Redis session-mode work should not need redis.exceptions at all.

Behavior preserved

• All signatures and return values are unchanged.
• Absence stays distinct from failure: validate_session_token still returns False for a
  missing token and raises only when Redis actually fails; from_redis still returns None
  for a missing record.
• RedisException() still constructs with no arguments.
• functools.wraps keeps names and docstrings intact for the static lane.

Behavior change (intentional, called out in CHANGELOG)

store_session_token previously caught any Exception and re-raised a bare
RedisException. It now translates only driver failures, so programmer errors propagate
unchanged instead of being masked. The error log line it emitted is preserved.

Tests

New test/test_redis_error_contract.py (164 tests) injects representative connection, timeout,
response, busy-loading and cluster failures into every affected public operation and asserts
RedisException. It also covers cause chaining, non-translation of programmer errors,
unchanged success paths and lifecycle, the absence-vs-failure distinction, and a guard test
that fails if a new RedisBase public method is added without coverage. The tests are
mock-based and need no running redis-server.

165 passed   # new suite + test_misc.py::test_changelog_consistency

flake8 is clean on dcicutils/redis_utils.py, dcicutils/redis_tools.py, and the new test file.
The existing redisdb-backed suites were not run locally: redis_exec in pyproject.toml points
at /usr/local/bin/redis-server, which is not the homebrew prefix on arm64. That is a
pre-existing environment gap and untouched here; CI runs those.

Versioning

Bumped to 8.20.0 (added public symbols) with a matching CHANGELOG.rst entry, per
test_misc.py::test_changelog_consistency.

[coveralls]

Coverage Report for CI Build 31202017610

Coverage increased (+0.1%) to 74.762%

Details

• Coverage increased (+0.1%) from the base build.
• Patch coverage: 33 of 33 lines across 2 files are fully covered (100%).
• No coverage regressions found.

Uncovered Changes

No uncovered changes found.

Coverage Regressions

No coverage regressions found.

---

Coverage Stats

Relevant Lines:	15643
Covered Lines:	11695
Line Coverage:	74.76%
Coverage Strength:	0.75 hits per line

---

💛 - Coveralls