fix(build): default to a Release build type

[youtalk]

What

Default CMAKE_BUILD_TYPE to Release when the user has not chosen one, leave any explicit -DCMAKE_BUILD_TYPE alone, and offer the four standard types to cmake-gui / ccmake.

Why

The project sets no build type and no optimisation flag anywhere, and neither docker/Dockerfile nor docker/Dockerfile.cpu passes one. An unset build type is not a neutral default: CMake then emits no -O flag at all, so every binary built from this repo, the released container images included, is compiled -O0.

That is measurable, not theoretical. Disassembling a shipped aarch64 binary, the per-frame input tensor normalisation in modules/models carried not one vector instruction, reloaded MEAN and STD from memory on every iteration, and issued a scalar fdiv per element:

ldr  s1, [x0]              ; src[i]
ldr  s0, [x0, x1, lsl #2]  ; MEAN[c]  -- reloaded every iteration
fsub s1, s1, s0
ldr  s0, [x0, x1, lsl #2]  ; STD[c]   -- reloaded every iteration
fdiv s0, s1, s0            ; a division, per element
str  s0, [x0]

objdump | grep -cE "\.4s|\.8h|\.16b" over that whole function returned 0.

Effect, measured

On an R-Car X5H (Cortex-A720), one clip, 198 inferred frames, same source built the same way apart from this commit:

	input prep	frame wall	fps
as shipped (no build type, -O0)	33.8 ms	57.3 ms	17.5
with this commit (Release)	13.9 ms	37.0 ms	27.1

Per 1024x512 image, that one conversion goes from 14.4 ms to 6.4 ms, and it runs three times a frame. Nothing here is board-specific — the affected code is plain CPU work on the shared pipeline, so an x86_64 host running the CUDA or TensorRT provider was compiled -O0 too.

Why Release and not RelWithDebInfo

Deliberate, and the difference is not cosmetic. -O2 leaves that conversion completely unvectorised — measured 0 vector instructions in the loop at -O2, against a NEON ld3 de-interleaving loop at -O3. Only Release gets the win.

Risk

Release adds -DNDEBUG, which disables assert(). Checked before choosing it: there are no non-static_assert assert() calls anywhere under modules/ or app/, so nothing changes behaviourally.

Verification

Built docker/Dockerfile.cpu on this branch: the configure step reports Build type: Release, -O3 reaches the actual compile lines, the VisionPilot target links, and -DCMAKE_BUILD_TYPE=Debug is still respected.

Relation to the other two PRs

This is the first of three independent changes; the other two are opened alongside it and none of them depends on this one to build. Suggested review/merge order is this PR, then the input-tensor fusion, then the merged-graph NPU backend, because the later two are measured on top of a Release build.

Note on spell-check-differential

The check is red, and none of it comes from this branch. spell-check-differential scans every changed file in full, so it reports words that already live in those files on main — Ipopt, libnice, CPACK, cipo, latc, Matx, dets, and the forbidden ROS2 among them. The same check has been failing on main in spell-check-daily every night for at least a week, and the last two merged PRs (#408, #411) merged red on it too. Every word this branch actually introduces is either already in the shared dictionary or added to .cspell.json here, verified locally with cspell against the repo config.