Draft: add -m/--mask flag to dotenv list (sensitivity-gated)

[maorq08]

⚠️ Draft — not a contribution proposal

This is a draft PR opened for internal review only. It is not being proposed as an
official contribution to python-dotenv and is not ready to merge. Please do not review
or merge it as an upstream contribution.

What this adds

A -m/--mask flag on dotenv list that redacts values whose key looks sensitive,
so secrets don't get printed in plaintext by dotenv list (e.g. in a terminal recording,
CI log, or screen share).

$ dotenv list --mask
API_KEY=ab****gh
DB_PASSWORD=****
PORT=8080
HOST=example.com

Masking rule

Two independent pieces, both gated per-key:

1. Sensitivity heuristic — _is_sensitive_key(key) does a case-insensitive substring
   match against a fixed keyword list: KEY, SECRET, TOKEN, PASSWORD, PASSWD,
   PWD, CREDENTIAL, AUTH, PRIVATE, ACCESS, CERT, DSN, CONNECTION_STRING,
   CONN_STRING. This is the same substring-match convention used by mise-en-place,
   Apache Airflow's keyword-based masking, and gitleaks' keyword list — e.g. MY_API_KEY
   matches on KEY, DB_PASSWORD matches on PASSWORD. A key that doesn't match prints
   its value completely unchanged, even with --mask on.
2. Length-based partial reveal — _mask_value(value), applied only to values whose
   key matched the heuristic:
   • values longer than 4 characters keep the first and last 2 characters, e.g.
     MY_SECRET_KEY → MY****EY
   • values of 4 characters or fewer become **** entirely

None/unset values (a bare key with no =value in the .env file) are left untouched
either way, so --format=json's null output and the plain formats' existing
"skip if value is None" behavior are unchanged.

Format-agnostic by construction

The masking transform is applied once, to the values dict returned by
dotenv_values(stream=stream), immediately after that call and before the --format
branch in list_values (src/dotenv/cli.py). That means simple (default), json,
shell, and export output are all masked consistently — the transform isn't
duplicated per format branch.

Tests

Added to tests/test_cli.py, following the file's existing cli/dotenv_path fixture
and CliRunner-based style:

• Existing test_list_mask table extended across all four --format values (values

  4 chars, ≤4 chars, and a bare/None key).

• test_list_mask_non_sensitive_key_unmasked — the key case proving the heuristic
  actually gates masking rather than masking everything (PORT=8080 prints unmasked).
• test_list_mask_short_flag — covers the -m alias.
• test_list_mask_empty_value — an explicit KEY= (empty string, distinct from a bare
  key with no = at all) masks to ****.
• test_list_mask_five_char_boundary — a 5-character value is the first to cross into
  the partial-reveal branch.
• test_list_mask_custom_file — --mask combined with an explicit non-default
  --file path.
• test_list_mask_mixed_sensitive_and_non_sensitive (+ a --format=json variant) — a
  sensitive and a non-sensitive key in the same file, so per-key gating is visible in
  one result rather than only single-key cases.

Full suite: 266 passed, 1 skipped locally (plus 3 pre-existing failures in
test_run_with_invalid_cmd/test_run_with_version/test_run_with_command_flags that
reproduce identically on unmodified main — they're a macOS BSD-printenv environment
incompatibility, unrelated to this change). ruff check and mypy are clean on both
changed files.

CHANGELOG

Added an ## [Unreleased] / ### Added entry describing the sensitivity-gated behavior.

[maorq08]

Closing - this was a draft opened for internal review only, not an upstream contribution proposal.