Background agent messages now depend on whether its output file can still be read

What

• A new canReadOutputFile flag is now tracked on background/async agent launches, defaulting to true, and set explicitly in several agent dispatch and relaunch code paths.
• Status and notification messages for a stopped or interrupted background agent now check this flag: when it's false, Claude Code tells you to resume the agent and ask it for a status report, instead of suggesting you check its worktree or output file for partial work.
• The message shown when a background agent from a previous session can't be restarted now also depends on this flag: if the output file can be read, it still suggests checking the worktree/output; if not, it suggests messaging the agent for a status report instead.

Why Not every background agent leaves behind a readable output file. This change makes the guidance Claude Code gives you match what's actually possible, instead of pointing you to a file you can't inspect.