Add Slack user resolver tool for natural language names

[fbraza]

Summary

• expose a slack.user.resolve tool that fuzzy-matches spoken names to Slack users
• cache Slack users_list data with expiration to limit API calls
• return structured matches (user_id, real_name, confidence) for downstream tool usage

Acceptance Criteria

• resolver handles first/last/display names and ignores bots/deleted users
• ambiguous or low-confidence matches prompt clarification or return multiple options
• prompt instructions guide the LLM to resolve names before messaging tools

Notes

• integrate resolver internally as Python helper, expose as FunctionToolset tool
• document usage alongside slack.private.message
• consider partial-name fuzziness while avoiding false positives

Example

class SlackUserMatch(BaseModel):
    requested: str
    user_id: str
    real_name: str
    confidence: float

@slack_toolset.tool(name="slack.user.resolve")
def resolve_users(ctx: RunContext[Deps], params: SlackResolveParams) -> list[SlackUserMatch]:
    cache = ctx.state.setdefault("user_cache", SlackUserCache())
    users = cache.ensure(ctx.deps.client)

    matches = []
    for name in params.names:
        match = fuzzy_match(name, users)
        if match:
            matches.append(SlackUserMatch(requested=name, **match))
    return matches