df11pack v0.1.1

Fixes found by an independent review of the repository against its own data.

Fixed

• --luts=correct did not change the LUTs in 0.1.0. It wrote the
  df11pack_luts="correct" stamp but encoded in compat mode, so 0.1.0 output
  stamped "correct" is compat output (it decodes correctly; the stamp is wrong).
  The mode now reaches the encoder, with a test that checks the written LUTs.
• Two definition patterns claiming the same module are refused; before, the
  module became two units and its tensors were written twice.
• --ram with a huge value gives a parse error instead of overflowing.
• A unit whose tensors are all empty is an error, not a panic.
• verify fails a unit whose gaps/output_positions lack an entry for a window
  or chunk the stream enters (before, it skipped it).
• Opening a safetensors file with a malformed header (non-integer shapes or
  offsets, offsets past the data, sizes that don't match the dtype) fails at once,
  naming the tensor.
• --io auto finds the disk type for NVMe and other partitions, instead of
  "device type unknown" (the choice made was already right).

Changed

• verify on --index idx8 output exits 2 ("idx8 output: not supported by
  verify") instead of reporting every unit as FAIL with exit 1.
• compress says what it wrote ("wrote 5 safetensors files + config.json").
• --idx8-block requires --index idx8 and refuses 0 at parse time (exit 2).
• --ram help notes there is always at least one worker.
• Definition headers say where each definition comes from (19 from
  ComfyUI-DFloat11-Extended's pattern_dict.py, 16 from a published release's
  dfloat11_config); the ComfyUI flux, chroma and chroma-radiance sources
  record Extended commit 414506d instead of master (content unchanged).
• Tests fail instead of passing silently when a tracked fixture is missing or
  unreadable; only absent generated fixtures are skipped, with a SKIP line.
• CI builds the MSRV job with Rust 1.85.1 (the repo's toolchain file had
  overridden it, so 1.85 was never really tested before).
• Documentation brought up to date with the measurements (FINDINGS, DESIGN,
  PLAN, COMPATIBILITY, INDEX_SCHEMES); NORMS.md added.

Added

• Contributor documentation: CONTRIBUTING, Code of Conduct (Contributor Covenant
  2.1), SECURITY, issue and pull request templates, CITATION.cff, and
  docs/DEFINITIONS.md for the model definition format.
• phase0/make_tier0.py --src, and fixture regeneration steps in phase0/README.md.
• CI uses current GitHub actions (Node 24) and pins Ubuntu 24.04.

Known limits (in addition to 0.1.0's)

• H12 is unmeasured: the CPU decoder behind verify is sequential, and there is
  no GPU verification path.
• config.json keeps the source model's schema (for example rope_theta,
  torch_dtype, the source's transformers_version), so it differs from the
  official compressor's save_pretrained output.
• verify --level full peaked at 12.0 GiB of RAM on Qwen3-8B (6.6 GiB to compress).