fix(#1045): de-duplicate GRAVITY and the 128.0 fall terminal; remove afloat-stall disclosure

[djhenry]

Summary

Fixes #1045. physics.rs's header claimed to be the single source of truth for physics constants, but that was falsified by its own file:

• fall_damage() locally shadowed GRAVITY with its own const GRAVITY: f32 = 120.0; instead of reading the module's real one.
• The controller's fall-speed clamp (movement's private MAX_FALL = 128.0) and fall_damage's local TERMINAL = 128.0 shared both a name-shape and a numeric value despite being different quantities — a velocity clamp on the controller vs. a damage-curve clamp on a derived impact velocity — a same-name-different-quantity trap for anyone sweeping by name later.

This applies the issue's own preferred fix (option 1):

• Promoted the controller's terminal fall speed into physics.rs as FALL_TERMINAL_VELOCITY, with a doc comment explicitly warning it must not be confused with fall_damage's clamp.
• movement now imports FALL_TERMINAL_VELOCITY/GRAVITY from eqoxide_core::physics instead of defining its own MAX_FALL.
• Deleted fall_damage's local GRAVITY shadow entirely — it now uses the real module constant.
• Renamed fall_damage's local TERMINAL to IMPACT_VELOCITY_CAP, a name that can't be mistaken for the controller's terminal.
• Left eqoxide-nav::collision's unrelated MAX_FALL = 120.0 (a fall-search distance cap) untouched — the issue explicitly warns against conflating it with the other two, and it's a genuinely different constant used for a different purpose.

As a side effect, MAX_FALL as a name now exists exactly once in the whole repository (collision.rs's unrelated one), which resolves the issue's flagged "trap for anyone sweeping by name" concern.

Also: removed the afloat-stall disclosure (#776/#801) entirely

Bundled into this branch on review: the afloat_stall signal — "body afloat in water, driven to swim somewhere, making no net progress for 3.0s" — was purely observational (a throttled log line plus a player.afloat_stall field on GET /v1/observe/debug), with no engine/physics behavior depending on it. Unattended-AI-dev feature bloat, and splitting an affordance like this out per-medium (water only, nothing analogous on land) doesn't make sense as a permanent feature. Removed entirely:

• AfloatStall/AfloatStallClock/AfloatFrame and crates/eqoxide-core/src/afloat.rs outright.
• ControllerView's disclosures()/publish_disclosures()/invalidate_disclosures() API, collapsed to a single pub hold field with invalidate_hold().
• CharacterController::afloat_stall()/disclosures(), leaving just hold().
• All downstream call sites in GameState, the HTTP observe handler, the IPC/net publish paths, and app.rs's two render-frame publish sites.
• The dedicated test sections in eqoxide-core, eqoxide-ipc, eqoxide-http, eqoxide-net, and movement.rs, and the documented afloat_stall field from docs/http-api.md.

Test plan

• cargo check --workspace --all-targets --all-features — clean
• cargo clippy --workspace --all-targets --all-features — clean
• rbuild . test --workspace --locked — full workspace suite, 0 failed
• Swept the repo for stale MAX_FALL/TERMINAL references after the rename — none found outside the intentional collision.rs constant and unrelated "TERMINAL state" terminology
• Swept the repo for stale afloat/Afloat references after the removal — none found outside legitimately historical, unrelated Recovery::Afloat doc-comment prose (issues qcat spawn pocket: a swimmer at the pocket swim plane is mounted onto the ceiling lid and stranded dry (the actual cause of the #329 live wedge) #649/fix(#649): the depenetration push-out is water-blind — recover a swimmer at its own depth, not onto the nearest floor #658/qcat spawn pocket: the swimming step-up hauls a swimmer onto a tile floor 0.009 u above the waterline that it cannot get back off (the remaining #329 wedge) #661, a since-removed enum variant)

🤖 Generated with Claude Code

https://claude.ai/code/session_01CWrs7cDd1NAJmaLscBjgcj