fix: wait_exponential with zero multiplier no longer overflows into max wait

[TrueFurina]

Fixes #710.

Problem

With wait_exponential(multiplier=0) (or any zero multiplier), the result of multiplier * exp_base**exponent is always 0. But exp_base**exponent is computed first, and for large attempt numbers (2**1024 at attempt 1025 with the default exp_base=2) it raises OverflowError — which is caught by the except OverflowError: return self.max branch, so the strategy wrongly returns max instead of 0.

This is easy to hit with a long-lived @retry(wait=wait_exponential(multiplier=0)): the process retried hourly for a few days, then every "failure" wait silently became the maximum wait.

Fix

Short-circuit the zero-multiplier case before computing the power:

if self.multiplier == 0:
    result: float = 0.0
else:
    exp = self.exp_base**exponent
    result = self.multiplier * exp

The min clamp still applies afterwards, so wait_exponential(multiplier=0, min=5) correctly returns 5.

Testing

• pytest tests/test_tenacity.py -k exponential — 18 passed (16 existing + 2 new regression tests: test_exponential_zero_multiplier_returns_zero_for_huge_attempts covering attempts 2/10/1025 and exp_base=2.0, and test_exponential_zero_multiplier_with_min_wait)
• Normal-path and max-capping behavior unchanged (multiplier=1 still returns 1/2/4/8... and caps at max)

[TrueFurina]

Lint fix: the test (3.14, lint) job failure came from my touched test file - one hunk needed a line wrap and the import block needed sorting per the project ruff config (line-length 88). Both fixed and verified locally with the project's own ruff settings (format --check and check both exit clean). tenacity/wait.py needed no changes.

[TrueFurina]

Correction: my earlier lint fix used an incomplete copy of the project ruff config (the isort subsection was dropped), so my local run passed while CI correctly flagged I001 at tests/test_tenacity.py:16:1. Re-did the fix with the full [tool.ruff.lint] config including [lint.isort]; the import block is now ordered per the project settings and both ruff check and ruff format --check pass locally with the complete config.

[TrueFurina]

Root cause found via the re-triggered run: two real violations my earlier local check missed because I had reconstructed the ruff config by hand instead of using the project file - I001 (import block order, fixed via the project ruff config) and B904 on the intentional implicit-chaining raise in the new overflow test (marked # noqa: B904, since implicit __context__ chaining is exactly what that test exercises). Verified locally with the CI-pinned ruff 0.16.9 against the project's own pyproject.toml: ruff check and ruff format --check both exit clean.