docs: add technical report for PR #296

Covers the effective-affinity-mask reporting fix: the three-way contradiction
in the environment block under an inherited mask, why the signals disagreed
about whether they could see one, and the three secondary defects the branch
review turned up (locale-dependent readback, the mask being read from the tmux
server rather than the measured process, and nproc's OpenMP thread cap reaching
the printed core count).

Records two things the PR itself does not: that reading /proc/self/status
Cpus_allowed_list would have avoided the locale defect and the taskset-free
path outright, at the cost of a hotplug-window discrepancy worth measuring
first, and that this script still has no committed regression harness even
though an ad hoc one is what caught four defects here.

Bilingual, following the existing TECHNICAL_REPORTS layout. Force-added because
.gitignore excludes *.md.