Declare rustler explicitly so mdex_native force-builds

[timujinne]

MDEX_NATIVE_BUILD=1 forces mdex_native to build its NIF from source when no precompiled build exists for the running OTP (28 in our case), which requires rustler itself, not just rustler_precompiled. mdex arrives here transitively through phoenix_kit; a clean checkout fails to compile without this declaration. Same declaration phoenix_kit's own mix.exs already carries.

---

One fix, six repositories. Every module that pulls mdex transitively through phoenix_kit needs the same one-line declaration, and each repository needs its own PR. Already merged: BeamLabEU/phoenix_kit_catalogue#77, BeamLabEU/phoenix_kit_billing#27. Still open: BeamLabEU/phoenix_kit_crm#25, BeamLabEU/phoenix_kit_entities#35, BeamLabEU/phoenix_kit_document_creator#42.

Scope. This makes this repository build from a clean checkout under MDEX_NATIVE_BUILD=1. It does not remove the need for a host application to declare rustler itself: optional dependencies of dependencies are never resolved, so whichever project is the root has to declare it.

[timujinne]

Ran the full suite on this branch tip: mix deps.get, mix compile under MIX_ENV=test, then mix test with the :integration half enabled against a real Postgres, with MDEX_NATIVE_BUILD=1 set under OTP 28 — the scenario this PR's rustler declaration targets. The declaration compiles and builds the native NIF cleanly.

Note on running the integration half at all here: config/test.exs on this branch hardcodes phoenix_kit_warehouse_test<partition> as the database name with no env override. On a Postgres role without CREATEDB, mix test doesn't fail — it silently excludes the entire :integration half instead (80 tests, 0 failures, 701 excluded), which looks green without having exercised anything DB-backed. To get real coverage I temporarily pointed config/test.exs at an existing database via an env-driven override, ran the suite, then reverted the change before touching git (not part of this PR — no commit, no push).

With the integration half actually running: 781 tests, 10 failures.

All 10 predate this PR and are unrelated to it — this PR only adds {:rustler, ">= 0.0.0", optional: true} to mix.exs/mix.lock, a build-time dependency declaration that touches none of the failing code paths. I read each failure's stack trace individually; all trace back to the same root cause: phoenix_kit_catalogue (a hard dependency of this module) is published on Hex pinned to phoenix_kit ~> 2.8, while the core actually resolved and running here is 2.13.5 — a real gap between the core version catalogue was built/tested against and the one in use. That shows up as: an admin-tab count assertion off by one (37 expected vs 38 actual), a foreign-key constraint on performed_by_uuid that the test fixtures don't account for, add_source_ref/3 returning {:error, :invalid_ref_type} where the tests expect success, a permission-redirect check that no longer holds, and a stale comment-count assertion.

None of this is introduced by this PR, and I haven't touched any of it here.