[wxs-lang]

The existing suite (55 tests) covers one checksum example per barcode type, EAN-13's checksum/no-checksum interplay, GS1-128 symbol sequences, Code 128 build stability, and the writers. Several encoders and the lookup layer had no dedicated tests at all (UPC-A, ITF and Codabar only appeared in the visual test_manually.py). This PR adds 239 tests covering that missing pure-encoding logic; no production code is changed and no image rendering is asserted (only build() patterns and value-level behaviour).

• barcode/__init__.py registry (test_registry.py) — PROVIDED_BARCODES is exactly the 22 sorted names; every name resolves via get_barcode_class to its class; aliases (ean→ean13, gtin→ean14, isbn/gs1→isbn13, nw-7→codabar, upc→upca) resolve to the identical class object; lookup is case-insensitive; unknown names (including "" and names with stray whitespace) raise BarcodeNotFoundError with the message The barcode '<name>' is not known., with or without a code; get with a code returns an instance, without one the class, and options are passed through to the constructor
• EAN family (test_ean_family.py) — EAN-13: overlong input is trimmed to 12 digits and the checksum recomputed, too-short input raises NumberOfDigitsError, non-digits and empty input raise IllegalCharacterError, calculate_checksum(value) for explicit values, the exact 95-module build() pattern, guard-bar fullcode format, to_ascii() output; EAN-8 and EAN-14: checksums for a range of values (including all-zero and all-nine inputs), trimming, length/character errors, build lengths (67 / 102 modules), EAN-8 guard fullcode; JAN: country codes at the 450/459/490/499 boundaries accepted, 460/489/500 rejected with WrongCountryCodeError, and the country check running before the length check
• UPC-A (test_upc.py) — checksums for five values, 12-digit input being trimmed with the checksum recomputed (a wrong 12th digit is silently ignored), length/character errors, make_ean=True prefixing a 0 in str()/get_fullcode() while the stored code keeps 12 digits, the exact 95-module build() pattern, and to_ascii() using _ for spaces
• ITF (test_itf.py) — odd-length input padded with a leading zero, exact interleaved build() patterns for digit pairs, pattern length scaling with pair count, narrow/wide width scaling, and IllegalCharacterError for non-digit input (including empty and separator-containing strings)
• ISXN validation (test_isxn_validation.py) — ISBN-13: hyphen stripping, rejection of non-978/979 prefixes, the 979 fourth-digit rule raising the plain BarcodeError, accepted 979-1 and 979-8 prefixes; ISBN-10: the documented example, the supplied check digit being ignored and always recomputed, hyphen stripping, and X check digits for four values; ISSN: X and numeric check digits, conversion to EAN-13, hyphen handling, and make_ean() output
• Code 39 and PZN (test_code39_pzn.py) — appended checksum characters for a range of inputs, upper-casing of lowercase input, add_checksum=False, building every one of the 43 legal characters framed by the start/stop pattern, illegal-character error paths, and PZN length validation, non-digit rejection, trimming, the checksum-10 rejection, and PZN8's seven-digit form
• Codabar (test_codabar.py) — exact build() patterns at default and custom narrow/wide widths, all six special characters ($:/.+-) building, and validation being deferred to build(): bad start/end characters raise BarcodeError, illegal body characters raise IllegalCharacterError, while the constructor accepts anything
• Code 128 charsets (test_code128_charsets.py) — charset A/B dictionary values and sizes, the charset C switch table (TO_B 100, TO_A 101, FNC1 102), start/switch tables, and encoded symbol sequences for digit pairs, start-code folding to START_B, a trailing odd digit switching to charset B, and control characters switching to charset A; plus the illegal-character error
• Charset table invariants (test_charset_tables.py) — EAN LEFT_PATTERN shape, C patterns being the bitwise complement of A and B the reverse of C, UPC tables being identical to the EAN A/C tables, ITF's two-of-five property (exactly two wide elements per digit), Code 39 table/map shape and sequential values, Code 128's 106 unique 11-module patterns, and the Codabar table shape
• Errors and base class (test_errors.py) — BarcodeError message storage/rendering, the four specific errors subclassing BarcodeError, the base Barcode raising NotImplementedError on instantiation, and the default writer being SVGWriter

Behaviour pinned as-is (possible upstream quirks, documented in the tests rather than "fixed"):

• Code39 with an illegal character raises KeyError (not IllegalCharacterError) under the default add_checksum=True, because the checksum is computed before character validation; with add_checksum=False the proper IllegalCharacterError is raised
• Code39.calculate_checksum() sums self.code including the already-appended checksum character, so calling it after construction returns a different character than the one in the code
• Code39("") silently produces the code "0"
• JAN with a non-numeric prefix raises a plain ValueError from int(), not a BarcodeError
• ISSN performs no length validation: ISSN("12345") silently produces issn == "123455" and a meaningless EAN-13 (9771234550029)
• ISSN.make_ean() returns the check digit verbatim, so it can contain X (e.g. 977114487500X); the numeric EAN checksum is only applied by the EAN-13 layer
• The ISBN-13 979 check accepts a fourth digit of 1 or 8, although the error message only mentions 97910 and 97911
• Code128("") raises IndexError (in _try_to_optimize), and CODABAR("") raises IndexError on build()

Verification (Python 3.12, package installed with pip install -e ".[images]", plus pytest and pytest-cov; no new conftest is needed — the tests live in the existing tests/ package and run from the repository root):

$ python -m pytest tests/test_registry.py tests/test_ean_family.py tests/test_upc.py tests/test_itf.py tests/test_isxn_validation.py tests/test_code39_pzn.py tests/test_codabar.py tests/test_code128_charsets.py tests/test_charset_tables.py tests/test_errors.py -q
239 passed
$ python -m pytest tests/ -q   # full suite, repository's own addopts (with coverage)
294 passed   # 55 pre-existing + 239 new