fix(pt_BR): allow repeated digits in numeric CNPJ roots

[liuyaohui666]

What does this change

Numeric Brazilian CNPJ generation now samples the eight root digits with replacement. A regression test uses the Receita Federal example 18.781.203/0001-28 to prove that repeated root digits are supported while preserving checksum generation.

What was wrong

_company_id_base(False) used random_sample(range(10), 8), which samples without replacement. That made every generated eight-digit root contain eight distinct digits, although valid numeric CNPJs may repeat digits.

As a result, Faker could generate only 10P8 = 1,814,400 roots out of the 10^8 = 100,000,000 possible numeric roots (about 1.8%).

How this fixes it

Use random_choices(range(10), length=8) so each root position is sampled independently. The alphanumeric path, branch suffix, public API, and checksum calculation remain unchanged.

Fixes #2445

Verification

• pytest tests/providers/test_company.py -q: 93 passed
• tox -e py -- tests/providers/test_company.py: 93 passed, then 101 passed in the locale module run
• Black (line length 120), isort, flake8, mypy, check-manifest, and test-class ordering checks passed
• git diff --check passed

AI Assistance Disclosure (REQUIRED)

• No AI tools were used in the creation of this PR.
• If AI tools were used, disclose how they were used and how you reviewed their output.

OpenAI Codex (GPT-5) assisted with repository/source-history inspection, duplicate searching, the reproducer, the regression test, the one-line implementation, and drafting this PR. I reviewed the diff and ran the listed tests and static checks locally.

Checklist

• I have read the CONTRIBUTING document.
• I have read the Coding Style documentation.
• I have run make lint.

make is unavailable in the Windows development environment. Its Black, isort, flake8, and mypy components were run individually and passed; the complete GitHub Actions matrix is left to verify the repository-native workflow.