fix(demos): cap build parallelism to avoid gateway OOM

[mfaferek93]

Problem

Docker builds for the demos fail intermittently with:

c++: fatal error: Killed signal terminated program cc1plus
... ResourceExhausted: cannot allocate memory

The build always dies in ros2_medkit_gateway. It is not a compile error and not architecture-specific (diagnostic_bridge and the other packages build cleanly). It is an out-of-memory kill of the compiler.

Root cause

The gateway has a few heavily-templated translation units (HTTP handlers) that peak at ~1.6 GB RSS each when compiling. colcon/make compiled them in parallel, one job per CPU core, so total peak scaled with core count. On many-core or RAM-limited Docker hosts (e.g. Docker Desktop), several heavy units overlap and exceed the memory limit, so cc1plus is OOM-killed. Whether a given machine succeeds depended on core count and scheduling timing, so it was flaky (passed on 8-core/8 GB, failed on a higher-core host even at 12 GB).

Fix

Limit build parallelism in the four demos that build the gateway from source: --executor sequential (one package at a time) + MAKEFLAGS='-j 2' (at most two compilers at once). Peak drops to ~3.2 GB, making the build deterministic across hosts.

manymove_industrial already has its own parallelism knob (MANYMOVE_COLCON_WORKERS) and is left unchanged.

Verification

Built the gateway from main inside a container hard-capped at --memory=6g --memory-swap=6g:

Finished <<< ros2_medkit_gateway [2min 19s]
Summary: 4 packages finished

Passes at 6 GB with margin; the previous unbounded build was OOM-killed under the same cap.