fix: prevent None corruption in agent state when outputs_to_state source key is absent

[IgnazioDS]

Summary

Fixes #10981.

When a tool's result dict does not contain a key referenced by outputs_to_state, result.get(source_key) silently returns None. That None then flows into state.set(), and when the state field uses the merge_lists handler it gets wrapped into [None], corrupting the state list with a spurious None entry.

Root cause (two layers)

1. _merge_tool_outputs (haystack/components/tools/tool_invoker.py) — no guard on output_value is None before calling state.set().
2. merge_lists (haystack/components/agents/state/state_utils.py) — new if isinstance(new, list) else [new] wraps None into [None] instead of treating absent output as a no-op.

Fix

tool_invoker.py — skip state.set() when the tool didn't produce the mapped output:

output_value = result.get(source_key) if source_key else result
# Skip state update when the tool didn't produce this output key
if output_value is None:
    continue
state.set(state_key, output_value, handler_override=config.get("handler"))

state_utils.py — treat None as "nothing to merge":

if new is None:
    return current_list

Both layers are needed: the tool_invoker fix is the primary guard; the merge_lists fix makes the function robust as a standalone utility.

Test plan

• Reproduce with a tool whose outputs_to_state maps a key that may be absent from the result dict — confirm state no longer contains None entries
• Existing merge_lists tests pass
• merge_lists(["a"], None) returns ["a"] (not ["a", None])
• merge_lists(None, None) returns []

[vercel]

@IgnazioDS is attempting to deploy a commit to the deepset Team on Vercel.

A member of the Team first needs to authorize it.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[sjrl]

@IgnazioDS thanks for jumping on this so quickly, but this was an issue we wanted to handle internally and did so in this PR #10990 so I'll be closing this PR