What changed

When your headphones are connected (you hear audio) but CAPod's dashboard shows no connection, your phone's Bluetooth stack is usually dropping the broadcasts CAPod relies on — common on some Xiaomi/HyperOS ROMs (#603). Two improvements:

• The dashboard now shows a hint pointing to the Troubleshooter in exactly that situation, instead of leaving it buried in Settings → Support.
• The Troubleshooter now applies the minimal compatibility tweak that makes reception work, and only changes your settings when it actually succeeds — no leftover options toggled on, and nothing changed after a failed or cancelled run.

Technical Context

• Probing now goes through a transient in-memory override on BlePodMonitor instead of writing the three compat settings on every attempt; only the winning combo is persisted, and only on terminal success. Failure/cancellation clears the override, restoring untouched settings.
• Combos are tried fewest-disables-first, so a phone that only needs batching disabled isn't also left with filtering disabled (ongoing CPU cost).
• Each probe clears the device cache and counts only observations past a settle cutoff, so a prior combo's cached devices (kept ~20s) can't be mistaken for the current one working. "Found" checks require a fresh live BLE observation, not cached/AAP state — so it can't exit early with "nothing to troubleshoot" for the devices that need it.
• The dashboard hint is debounced ~15s and suppressed whenever any pod is live, so it never flashes during connect→first-broadcast and never claims "no data" while data is arriving.
• Known limitation: during the ~1.5s scanner-restart window a lingering old-combo case-context frame can still mask a fresh non-case one via preferCaseContextPod — a rare, self-correcting false-negative left as-is to avoid touching shared monitoring behavior.
• Scope: addresses the discoverability/robustness half of BLE broadcasts only received during pairing on Xiaomi Redmi Note 13 5G  #603. The separate "device shows twice when hardware batching is off" duplicate needs a verbose repro log and is not in this PR — hence Refs, not Closes.

Review checklist

• In-memory override always cleared on every exit path (success, failure, cancellation, persist-throw)
• Only the minimal winning combo is persisted, only on terminal success
• collectFreshDevices() doesn't discard observations on timeout
• Dashboard hint stays hidden while any pod is live and during the debounce window

Refs #603