Implement the aba, verhoeff, and damm identifier validators

[jzonthemtn]

Description

The redaction policy schema names eleven validators for the custom identifier filter.
phileas/filters/identifier_validators/__init__.py registers eight: luhn, mod11, mod97,
mod23_letter, es_cif, de_steuerid, de_personalausweis and bic_structural. A policy naming
aba, verhoeff or damm raises at load, from resolve_validator in
phileas/filters/validators.py:90.

The spec describes them as:

Validator	Description
aba	ABA routing transit number checksum (weights 3-7-1 mod 10).
verhoeff	Verhoeff check digit (dihedral-group scheme).
damm	Damm check digit (quasigroup scheme).

All three ports had this gap. philterd/phileas-dotnet#89 closes it for .NET and is a working
reference; philterd/phileas#391 tracks it for Java.

Notes from implementing it in .NET

The tables and the tests are language-independent, so most of the work transfers:

• Published vectors that exercise each scheme. Verhoeff accepts 2363, 123451 and 758722 and
  rejects 12345. Damm accepts 5724 and 112946, rejects 5727 and 112949. ABA accepts
  021000021.
• Two structural properties worth asserting, because a transcribed lookup table can be subtly
  wrong and still look plausible. For any body exactly one appended digit may validate; and neither
  scheme may accept a value with two adjacent digits transposed, which is the property that
  distinguishes them from a plain modulus check. Assert both across a range of values, not on single
  examples.
• ABA should require exactly nine digits and reject a tenth rather than letting the weights wrap.

Two things specific to this port:

• \d is Unicode-aware here, as it is in .NET. re.fullmatch(r"\d{9}", "０２１００００２１")
  matches, so a routing number written in fullwidth digits would pass a naive \d-based check. Use
  [0-9], or re.ASCII. Java does not have this problem, so its implementation will not show it.
• The module docstring claims the validators "are parity ports: for the same input they return the
  same result as the Java implementation." Java does not have these three yet
  (philterd/phileas#391). If this lands first, that sentence needs qualifying rather than leaving a
  claim that is no longer true.

Registration is a register_validator(name, factory) call in a new module under
phileas/filters/identifier_validators/, plus one import line in that package's __init__.py.

Acceptance criteria

• aba, verhoeff and damm are implemented and registered, so a policy naming one loads.
• The message raised for an unknown name lists all eleven.
• Each validator has unit tests covering a value the checksum accepts and one it rejects.
• A test asserts that for any body exactly one appended check digit validates, for verhoeff and
  damm.
• A test asserts neither scheme accepts a value with two adjacent digits transposed.
• aba requires exactly nine digits, ignoring separators, and rejects both eight and ten.
• Digits are matched as ASCII, so a value written in fullwidth or other Unicode digits is not
  accepted.
• A test runs each validator through a policy on the custom identifier filter, so the check is
  shown to run during filtering rather than only to resolve by name.
• The validator documentation lists the three, alongside the eight already documented, and the
  parity claim in the package docstring is still accurate afterwards.

Related

• philterd/phileas-dotnet#89 closes the same gap for the .NET port, with a working implementation.
• philterd/phileas#391 tracks it for Java.