Add Copilot instructions for PR review

[kandrosov]

Why

This repository had no copilot-instructions.md, so Copilot reviewed its PRs with no idea what
this code is for or what goes wrong in it.

The file added here is written for PR review, not onboarding — the five existing files in the
ecosystem were generated by Copilot's repository-onboarding feature in November 2025, and are
being rewritten in the same way (cms-flaf/FLAF#310 and its companions): what a useful comment
looks like here, the invariants worth checking with the failure signature each produces, the rule
that documentation ships in the same PR, and — deliberately — what not to comment on, since
noise is an automated reviewer's main cost.

Framework-wide rules are not duplicated; the file points at FLAF/.github/copilot-instructions.md
for them.

Testing

Documentation only — no code change. Every path, filename and directory referenced was verified to
exist in the current tree, and the facts table is dated so the next reader knows how much to trust
it.

What it says

This repository produces the numbers that go in the paper, and a mistake here does not crash — it
yields a limit that is plausible and wrong, with nothing downstream to catch it. The file is built
around that:

• allow_* flags are safety switches, not conveniences. resolveNegativeBins rejects a
  negative integral before it ever checks whether individual bins are negative-within-error, so
  allow_negative_integral disables a real protection. A new or widened allow_* is called out
  as a finding unless the PR justifies it — the honest fix is usually more statistics or dropping
  the process, and an amc@NLO sample with too few events is a reason not to use it rather than a
  reason to allow a negative integral.
• Datacard entries: a bare string means Process(name, name) while a mapping carries hist_name,
  is_signal, param_values and channels — the wrong form changes what is read, not just the
  label. The card's eras: list is its scope. Histogram names must match what the analysis wrote.
• bin_opt/ changes alter sensitivity: signal-derived edges applied to backgrounds, an
  optimisation silently returning fewer bins, signalFractionForRelevantBins semantics.
• Reproducibility: results are compared across eras, channels and analyses, so a dependence on
  dictionary or filesystem ordering is a finding even when one run looks fine.

Documentation: this repository has no site of its own, so a user-visible change needs a companion
PR to HH_bbtautau/docs/stat_inference.md, HH_bbWW/docs/stat_inference.md or FLAF/docs/.