fix(eq): re-apply equalizer when the native audio session rotates

ExoPlayer assigns a new audio session id after transient audio-focus
interruptions (navigation prompts, radar warnings), leaving the
AndroidEqualizer attached to the dead session so playback resumed
without equalization until the next station switch. The session-id
listener now detects genuine rotations through a dedicated guard and
re-activates the equalizer with the current preset, gated on EQ
availability to stay clear of player teardown/rebuild.

openspec/specs/eq-audiofocus/spec.md

@@ -0,0 +1,99 @@
+# EQ Audio-Focus Re-Apply Specification
+
+## Purpose
+
+Keep the native Android equalizer attached and equalized after ExoPlayer
+rotates the audio session id mid-playback (audio-focus transitions, ducking by
+other apps). Today re-attachment only happens on explicit station switch, so
+music resumed after an interruption is silently un-equalized. Covers
+`PluriWaveAudioHandler` (`lib/servicios/servicio_audio.dart`) only.
+
+## Requirements
+
+### Requirement: Session Id Rotation Triggers EQ Re-Apply
+
+The system MUST re-attach the equalizer and re-push the current preset's gains
+whenever `androidAudioSessionIdStream` emits a session id that differs from the
+last one processed for EQ purposes, is non-null, and `_eqDisponible == true`.
+
+Change tracking MUST use a dedicated field (`_ultimaSessionIdEq`), separate
+from the broadcast field `_androidAudioSessionId`, so existing broadcast
+semantics on `androidAudioSessionIdStream` stay unchanged.
+
+#### Scenario: Session id rotates while playing
+
+- GIVEN the player is playing, EQ attached/enabled (`_eqDisponible == true`),
+  a non-flat preset applied
+- WHEN `androidAudioSessionIdStream` emits a new non-null id different from
+  `_ultimaSessionIdEq`
+- THEN the system re-attaches the EQ (re-probes `_eq.parameters`, re-enables
+  per `_ecualizadorActivo`) and re-applies the current preset's gains
+- AND `_ultimaSessionIdEq` updates to the new id
+
+#### Scenario: Same id re-emitted produces no redundant re-apply
+
+- GIVEN `_ultimaSessionIdEq` already equals the last processed id
+- WHEN the same id is emitted again
+- THEN the system MUST NOT call the re-attach/re-apply choke-point again
+
+#### Scenario: First legitimate activation is not double-applied
+
+- GIVEN a station switch just completed `_recrearPlayer()` and its own
+  `_activarEcualizador()` call (`servicio_audio.dart:449`)
+- WHEN the resulting first session-id emission for the new player reaches this
+  listener
+- THEN the listener MAY invoke the same idempotent choke-point, but this MUST
+  NOT produce a different final gain state than the station-switch path alone
+
+### Requirement: Re-Apply Is Gated On EQ Availability
+
+The system MUST NOT attempt to re-attach or re-apply gains while
+`_eqDisponible == false`, to avoid racing `_recrearPlayer()`'s teardown/rebuild.
+
+#### Scenario: Rotation during player teardown is safely skipped
+
+- GIVEN `_recrearPlayer()` has set `_eqDisponible = false`, nulled
+  `_androidAudioSessionId`, and created a fresh `AndroidEqualizer()`
+- WHEN a session-id emission reaches the listener while `_eqDisponible` is
+  still `false`
+- THEN the system skips the re-apply choke-point entirely
+- AND no exception is thrown
+
+### Requirement: Re-Apply Is Independent Of Multi-Device Toggle
+
+The system MUST re-apply gains on session id rotation regardless of
+`_eqMultiDeviceEnabled`, since this fix lives below the `EstadoEcualizador`
+multi-device layer.
+
+#### Scenario: Re-apply works with the toggle on or off
+
+- GIVEN either state of `_eqMultiDeviceEnabled`
+- WHEN a session id rotates under the "rotates while playing" preconditions
+- THEN the re-apply behaves identically in both states
+
+## Non-Goals
+
+- Station-switch EQ path (`_cambiarFuente` → `_recrearPlayer` →
+  `_activarEcualizador()` at line 449) is unchanged.
+- Manual slider path (`cambiarBanda`/`setBanda`) is unchanged.
+- No new localized strings — no l10n keys added, removed, or modified.
+
+## Testability Matrix
+
+| Scenario | Testable how |
+|---|---|
+| Rotates while playing / same-id guard / first-activation no double-apply / teardown-race skip / toggle-independence | **Manual QA only.** `PluriWaveAudioHandler` is never instantiated anywhere in `test/` — confirmed zero matches for `PluriWaveAudioHandler(`. Its `_player` is a real `just_audio.AudioPlayer` requiring platform MethodChannels; `androidAudioSessionIdStream` is fed directly from it, with no fake/injectable seam. Verify on-device: real ducking (e.g. Google Maps prompt) during playback, and rapid station-switch during a ducking event for the teardown race. |
+| Station-switch path / manual slider path unchanged | **Regression only, not targeted unit tests.** Same instantiation blocker applies to both call sites; confidence comes from the full `flutter test` suite plus manual QA, not new automated tests against these paths. |
+
+**Correction to proposal**: the proposal's downstream notes claim
+`servicio_audio_source_switch_test.dart` shows "fake/stream patterns" for this
+handler. Verified against the file: it explicitly states *"We cannot
+instantiate the handler in unit tests (MethodChannels)"* and only exercises
+`ControladorReconexion` in isolation — not `PluriWaveAudioHandler`. No
+fake/injectable `AudioPlayer`/`AndroidEqualizer` seam exists in `test/`. Every
+scenario above is therefore manual on-device QA under the current harness, not
+automated. This reframes the proposal's Success Criteria item "covered by
+passing tests" as "passing tests where the harness allows, manual QA
+otherwise" — `sdd-design`/`sdd-tasks` should decide whether to accept that, or
+add a DI seam for `_player`/`_eq` as a separate prerequisite (out of this
+change's scope).