fix(cron): apply custom assistant rules in scheduled runs

[mabobsa]

Problem

Running a custom assistant via a scheduled task (cron) does not apply the assistant's rules (its system prompt). The agent receives only the bare task text, so it goes off-task and narrates in English instead of following the configured persona/language. Interactive runs are unaffected (the frontend supplies preset_context).

Root cause

Two independent gaps, both on the cron path:

1. Rule dispatcher not wired. build_cron_state constructs its own ConversationService and wires every assistant repo except the rule dispatcher — it never calls with_assistant_dispatcher. So resolve_assistant_snapshot finds assistant_dispatcher() == None, never calls read_rule, and rules_content stays empty. (The interactive build_conversation_state does wire it.)

2. Locale lost → rule file missed. Cron creates the conversation with assistant: None, so no locale reaches read_rule(id, None). That looks up {id}.md, but user rules are stored locale-suffixed as {id}.{locale}.md, so the file is never found even once the dispatcher is present.

Both must be fixed: without (1) read_rule is never called; with (1) but without (2) it returns empty for locale-less callers.

Fix

• Wire the assistant rule dispatcher into the cron ConversationService at the assembly site (build_module_states), mirroring the interactive path. build_cron_state's signature is left unchanged.
• read_rule falls back to any saved {id}.*.md rule file when the locale-specific file is absent, so locale-less callers (cron) still resolve rules.
• Adds a regression test (read_rule_user_falls_back_to_saved_locale_when_locale_missing).

Verification

• cargo test -p aionui-assistant — read_rule tests pass.
• End-to-end: a scheduled custom-assistant run now resolves non-empty rules, and the agent's [Assistant Rules] block contains the configured rules (confirmed via runtime logs and the stored conversation snapshot).