fix: route cache-write failures through exception_handler

[waltzforvenus]

Problem

exception_handler currently only protects the cache read: client.get(key) is wrapped in try/except, but on a cache miss the write path is not. If Redis becomes unavailable between the failed get and the subsequent store (or the store itself fails — e.g. connection drop, OOM, cluster failover), the exception propagates to the caller even though an exception_handler was configured to make caching failures non-fatal.

Fix

Wrap the cache-miss path (calling the original function, serializing the result, and the Lua cache-set script) in a try/except that delegates to exception_handler when one is configured, matching the existing behaviour of the read path. Without a handler, the exception is re-raised, so behaviour is unchanged for users who don't use exception_handler.

elif not exception_handled:
    try:
        parsed_result = fn(*args, **kwargs)
        result_serialized = self.serializer(parsed_result)
        get_cache_lua_fn(self.client)(keys=[key, self.keys_key], args=[result_serialized, self.ttl, self.limit])
    except Exception as e:
        if self.exception_handler:
            parsed_result = self.exception_handler(e, self.original_fn, args, kwargs)
        else:
            raise e

The handler receives the same (exception, original_fn, args, kwargs) signature as on the read path, so existing handlers (which typically fall back to calling the function directly) work unchanged.

🤖 Generated with Claude Code

[waltzforvenus]

(sorry finally could be fucked to fix the formatting issues using claude, this is the version i use on prod)