[NaySurGithub]

Problem

Since #64, standing at the edge of a bridge or on top of a pillar and looking down past the edge ("traditional" Bedrock bridging) places nothing and shows no outline. The look ray hits no block within reach, so Perlin had nothing to target.

#64 removed Perlin's reach-around (Bridging::FindTarget) because it sent an invented click (pinned to the side-face edge, y = 0.999, up to 17° off the look ray), which got the account banned. The vanilla client does have its own reach-around, though, and that is what this PR ports.

What the vanilla client does

The behaviour below comes from IDA on the Windows client 1.26.60.5. The frame pick (0x144E090C0) builds the HitResult that ClientInstance::_tickBuildAction (0x1465A2950) then uses:

1. The normal pick hits no block (HitResult type 2 / 3), in the regular aiming mode (the pick context's byte 0).
2. If the look direction's y < -0.7 (0x144E0A75C, about 44.4° below the horizon), it clips blocks only, no liquids, from the eye to eye + (0, -2, 0) (xmmword_14F0AE7C0, BlockSource vfunc +0x198).
3. If that clip hits a block (type 0 / 1), it becomes the frame's hit:
   • it keeps its hit point mPos, on top of the block under the player;
   • the face is replaced by the look's dominant horizontal axis: (dir.z < 0) ^ 3 → 2 / 3, or (dir.x < 0) ^ 5 → 4 / 5 when |dir.x| > |dir.z|;
   • it sets the byte at +0x84.
4. Held repeats (0x14376B8D0) only use a direct hit when type == 0 && !hit[0x84]. A look-down hit therefore only serves the first press; an existing build line keeps going as before.

Because it is the frame's hit, the outline shows on that block. The build lambda (0x143771570) sends clickPos = mPos - BlockPos = (eye fraction x, 1, eye fraction z), with the horizontal face.

Change

• Perlin::UpdateAim: when the normal raycast hits nothing and look.y < -0.7, raycast blocks only from the eye, straight down, 2 blocks. A block hit becomes aimHit, with the vanilla face rule and buildAssist = true. The hit point is kept, so the outline shows and clickOf gives the vanilla click.
• AimHit::buildAssist / heldbuild::Hit::assist: HeldBuild::Next skips the direct-hit branch for these hits, like the !hit[0x84] check.

Nothing from the old FindTarget comes back: no body-footprint search, no 1.5-block support search, no invented edge click, and nothing happens above the -0.7 threshold.

Testing

macOS arm64 Release build, packet capture while bridging on a server:

Placement	Face	clickPos	Source
(1, 86, 16) looking north at 69.9°	2	(0.475, 1, 0.494)	look-down pick
(1, 68, 1) looking west at 68.6°	4	(0.572, 1, 0.652)	look-down pick
(1, 86, 15) 318 ms later, still held, moving north	2	(0.44, 1, 0.651)	held repeat: last built block + movement face (0x14376B8D0)
(1, 86, 9) / (1, 86, 8) / (1, 86, 7)	2	(0, 0, 0)	build line with no hit, 180 ms catch-up

• Look-down picks show the outline and place the block in front.
• No kick or disconnect during the session.
• Not checked: the touch-input pick path, which vanilla handles separately. This PR applies the look-down pick in every input mode.