build: list instance_constraints.hpp, unbreaking configure on master

[Yaraslaut]

master is currently broken — this unbreaks it

cmake configure fails on every job and platform on master:

CMake Error at CMakeLists.txt:321 (message):
  morph: these public headers belong to no target's FILE_SET HEADERS, so
  VERIFY_INTERFACE_HEADER_SETS never compiles them standalone and would keep
  passing if they stopped being self-contained (morph#230):

      include/morph/forms/instance_constraints.hpp

Observed on #263, where all 21 jobs fail at the Configure step, and reproduced
locally against origin/master.

Cause: two correct changes that collided

• forms: let per-instance data reach the framework's own schema keys (part of #164) #248 added include/morph/forms/instance_constraints.hpp
• build: verify every public header standalone, and fail configure if one is unlisted #265 added a guard requiring every public header to be in some target's
  FILE_SET

Each was green. Neither PR's CI saw the other, because each was tested against a
master that did not yet contain the other.

The guard is not wrong here. It caught precisely what it exists to catch: a
public header that VERIFY_INTERFACE_HEADER_SETS would never have compiled
standalone, while continuing to report success. The fix is the one the error
message asks for — list the header.

Verified

• configure succeeds again on master + this change
• the verify target now builds instance_constraints.hpp.cxx.o among its
  objects, so the header is genuinely checked rather than the guard merely
  silenced
• MORPH_CLIENT_ONLY=ON and QT + OFFLINE_SQLITE + NET both configure cleanly

Note for the next header

This failure mode is inherent to a configure-time guard evaluated against the
merged tree: a PR adding a header is green until the guard is on master, and the
guard is green until the header is. Nothing detects the pair before they meet, so
it surfaces on the first merge after either lands.

That's the cost of the guard catching real omissions, and I'd argue it's the
right trade — but it means the first new header after #265 was always going to
hit this once. This is that once.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!