Additional context matching logic for LEA multi-codec

When we match with the remote capabilities, we don't filter by
preferred audio context, since they are suggestion.

Instead, we populate settings into 2 separated vector:
- Matched settings with matched preferred context,
- Matched settings,

We then filter by the first vector, then by the second vector.

Bug: 306225778
Test: atest VtsHalBluetoothAudioTargetTest
Change-Id: I03f322e8de509127d218a9de6b41dd39b2ebdcba