fix(hamilton): resolve trash_core96 from the deck instead of a stale pointer

[WillVT84]

Problem

HamiltonSTARDeck._trash96 is assigned only in __init__, and get_trash_area96() returns it directly. That cached reference does not survive the deck being rebuilt or the trash being unassigned, which produces two distinct failures.

1. Deserialized decks raise even though the trash is present. serialize() encodes the 96 trash as a child resource and emits with_trash96=False. A deck restored through Deck.deserialize() therefore runs __init__ with the flag off and leaves _trash96 as None, while trash_core96 is re-attached from the serialized children. The resource is on the deck and correctly positioned, but the getter raises:

deck2 = Deck.deserialize(STARLetDeck().serialize())
"trash_core96" in [c.name for c in deck2.children]  # True
deck2.get_trash_area96()
# RuntimeError: Trash area for 96-well plates was not created. Initialize with `with_trash96=True`.

Anyone loading a saved layout has to re-establish the pointer by hand.

2. clear(include_trash=True) leaves a stale pointer. clear() unassigns the trash but never clears the attribute, so the getter hands back an orphaned Trash with parent = None. That fails later and further from the cause than an outright error would.

Fix

Resolve the resource by name against the child tree, mirroring how the base class's Deck.get_trash_area() already works, and only trust the cached reference while it is still assigned to this deck. The parent is self guard is what fixes the second case: an unassigned trash falls through to the lookup and then raises honestly.

Behaviour when the trash genuinely was not created (with_trash96=False) is unchanged.

Tests

Two regression tests in hamilton_deck_tests.py, one per failure mode. Both were confirmed to fail against the unfixed source — the round-trip test errors with the original RuntimeError, and the include_trash test fails with "RuntimeError not raised" — and to pass with the fix.

Verification

• pytest pylabrobot/resources — 225 passed. The 5 failures are pre-existing Opentrons deck tests that fetch labware over the network and hit SSL: CERTIFICATE_VERIFY_FAILED locally; identical on baseline.
• ruff check / ruff format --check (0.15.4, per the pin) — clean on both changed files.
• mypy (1.18.2, per the pin) — Success: no issues found in 2 source files.

Note

Not addressed here, but the underlying design smell is that serialize() writes with_trash/with_trash96/core_grippers as False/None because the data lives in the children — the "not very pretty to have this key though" comments. That constructor-flags-vs-children mismatch is what produced this bug, and may affect other cached references the same way. Happy to follow up separately if that's of interest.

🤖 Generated with Claude Code

[WillVT84]

@rickwierenga This may be an artifact of how we manage decks in our implementation, but I've been having to evoke a helper function to re-establish this pointer whenever we load our deck in a protocol. Figured I'd submit a bug fix here.