pt_BR numeric CNPJ roots never contain repeated digits

* Faker version: 40.36.0 and current `master` (`f33e820`)
* OS: Windows 11, Python 3.14.6

The `pt_BR` company provider builds the eight-digit numeric CNPJ root with `random_sample(range(10), 8)`. Because sampling is without replacement, every generated root has eight distinct digits. Valid numeric CNPJ roots can contain repeated digits; for example, Receita Federal's documented numeric example uses root `18781203`, which repeats both `1` and `8`.

This does not make the generated checksum invalid, but it unnecessarily restricts Faker to `10P8 = 1,814,400` roots instead of `10^8 = 100,000,000` possible eight-digit numeric roots—about 1.8% of the expected space.

Official reference: https://www.gov.br/receitafederal/pt-br/centrais-de-conteudo/publicacoes/documentos-tecnicos/sped/documentos-tecnicos-e-financeira/leiautes/anexo-ii-leiautes-gerais.pdf

### Steps to reproduce

```python
from faker import Faker

fake = Faker("pt_BR")
roots = [fake.company_id()[:8] for _ in range(10_000)]

print(sum(len(set(root)) < 8 for root in roots))
print(all(len(set(root)) == 8 for root in roots))
```

Current output:

```text
0
True
```

### Expected behavior

Each position in a numeric CNPJ root should be sampled independently, allowing repeated digits. The provider can use `random_choices(range(10), length=8)`, matching the with-replacement behavior already used by the alphanumeric branch.

### Actual behavior

`random_sample(range(10), 8)` guarantees that all eight root digits are unique, excluding the overwhelming majority of otherwise valid numeric roots.

### AI assistance disclosure

This report was prepared with assistance from **OpenAI Codex (GPT-5)**. It was used to inspect the current implementation and its git history, search for duplicate issues and pull requests, reproduce the behavior on current `master`, calculate the reachable output space, and draft this report. The reproduction and source references were independently verified against the checked-out repository and the linked Receita Federal documentation.

Start at the pt_BR company provider entry point used by company_id(), and reproduce the behavior with the 10,000-root example in the issue. Replace the numeric root's without-replacement sampling with independent digit sampling, then verify that generated roots can contain repeated digits while company IDs remain valid.

Written by the indexing model from the issue text.