Reduce verbose polling output for background agent tasks

[bradygaster]

Problem

When Squad (or any agent) spawns a background task via the task tool with mode: background and then calls read_agent with wait: true, the CLI renders every polling attempt as a separate log line:

○ Read: General-purpose agent (Keaton: Triage issues #8 and #6)
  └ Waiting for result…

○ Read: General-purpose agent (Keaton: Triage issues #8 and #6)
  └ Waiting for result…

○ Read: General-purpose agent (Keaton: Triage issues #8 and #6)
  └ Waiting for result…

This produces 7+ identical lines of output for a single read_agent call that takes ~2 minutes. The user sees noise instead of a clean waiting state.

Expected Behavior

A single read_agent call with wait: true should render as one line that updates in place (or at most shows a spinner/progress indicator), not N repeated log entries. Something like:

○ Waiting for Keaton: Triage issues #8 and #6… (127s)
  └ Done

Or simply collapse the repeated polling into one entry.

Context

• This happens with read_agent(wait=true, timeout=300) — the CLI appears to be rendering each internal poll cycle as a separate tool output
• The agent description (e.g., Keaton: Triage issues #8 and #6) is already in the first line — repeating it adds no information
• For agents that take 2+ minutes, this can produce 10+ identical lines of output
• The actual result rendering at the end is fine — it's just the waiting state that's noisy