bug(gameplay): Liliana's Spoils does not present its top-five library choice

[ninthworld]

Problem

Casting Liliana's Spoils did not allow the player to look at or choose from the top five cards of their library. The card proceeded without presenting the hidden-information choice its effect requires.

Expected behavior

When Liliana's Spoils resolves for its controller:

1. That player privately sees the top five cards of their library (or all remaining cards if fewer than five remain).
2. They may choose an eligible black card from among those cards to put into their hand.
3. They may also choose no card.
4. The remaining looked-at cards are placed on the bottom of the library in a randomized order.
5. No other player learns the identities of cards that were only looked at.

Acceptance criteria

• Resolving Liliana's Spoils suspends resolution and presents its controller with the top-five selection prompt.
• The prompt contains exactly the top five cards, or the entire library when fewer than five remain.
• Only eligible black cards can be selected for the hand.
• Choosing no card is supported.
• The selected card moves to the player's hand.
• Every unselected card moves to the bottom of the library in a deterministic seeded random order.
• Other players' views do not expose the looked-at cards.
• Empty-library and no-eligible-card cases resolve without stalling.
• Engine, protocol/view projection, and client tests cover the complete prompt-and-answer path rather than only the card data.

Investigation notes

The failure could be in card authoring, pending-choice generation, private view projection, or client prompt rendering. Diagnose the broken layer before narrowing the fix; do not silently resolve the effect when a legal player choice is owed.

[ninthworld]

The failure is most likely due to no client-side UI being created to handle showing a custom set of cards to the player for them to select options from. The UI has sub-windows for showing the a graveyard, that needs to be reused to show these cards.

[ninthworld]

Test-coverage investigation

The failure is primarily a browser presentation/interaction coverage gap, rather than an absence of the underlying engine choice machinery.

What is already covered

• The card definition correctly authors the opponent discard followed by a look_at_top of five, taking up to one black card into hand.
• Engine tests for the shared choice machinery cover suspension, filtering, optional selection, moving the selected card, random bottoming of the remainder, deterministic RNG use, and no-match/no-stall behavior.
• Server/view tests cover player_choice → select_from_zone, min: 0, chooser-only revealed cards, and non-disclosure to opponents and spectators.
• Protocol and client unit tests parse gameview-choice.json and cover selection bookkeeping and answer serialization.

What was missing

There is no test that names or plays Liliana's Spoils itself. In particular, nothing jointly verifies its discard-before-look sequence, exact top five, authored black filter, optional no-card answer, short/empty library behavior, resumption, and final move to the graveyard.

More importantly, no Playwright test renders gameview-choice.json, clicks a revealed candidate, and submits the answer. The current client renders revealed cards under shown to you inside the side panel's Stack tab. It does not automatically open or focus that surface when the choice arrives. Depending on viewport and prior tab/panel state, the decision can therefore be hidden even though the wire data and interaction machinery exist.

Why this escaped

The relevant work was split across three changes:

1. feat(engine): a general player-choice prompt — search, scry, reveal, discard #604 added the engine/protocol choice mechanism and extensive Rust tests, but no browser rendering test for the choice fixture.
2. feat(client): rebuild the client to the prototype #699 rebuilt the client and retained revealed in the side panel, but its browser suite still did not exercise the choice fixture.
3. feat(engine): print the whole M19 base set, and 48 more of its cards #720 added Liliana's Spoils in a 48-card catalog tranche and relied on representative component tests instead of a card-specific playable-path test.

Each layer was tested separately, but no test crossed catalog → engine suspension → private view → browser presentation → submitted answer.

Regression coverage recommended for this issue

• Add an issue_773_... engine test using the bundled liliana_s_spoils definition that covers the full discard/look/answer/resume path, including decline, no eligible card, short library, and seeded random bottoming.
• Add a Playwright test driven by gameview-choice.json that proves the choice is immediately discoverable, candidates are visible and clickable, selection is highlighted, confirmation sends the exact advertised ID, and zero selection works.
• Exercise that browser test at a narrow viewport as well.
• Preferably add one thin real-server smoke path for a hidden-zone mid-resolution choice, protecting the complete integration seam.

A broader audit found additional catalog and browser-coverage gaps; those will be tracked in a separate issue referencing this one.