fix: keep power iteration relative error nonnegative

[pralav-25]

Describe your change

Power iteration can stop after its first iteration when the Rayleigh quotient is negative. The relative-error denominator currently uses the signed quotient, so a negative estimate produces a negative error that immediately satisfies the positive tolerance.

For example, power_iteration(np.diag([-4.0, 2.0]), np.ones(2)) returns an eigenvalue of -2.8 with a residual norm of 2.4. Dividing by the magnitude of the quotient keeps the relative error nonnegative and allows convergence to the dominant eigenvalue, -4.

• Add an algorithm?
• Fix a bug or typo in an existing algorithm?
• Add or change doctests? -- This PR changes code only, following the repository's request to keep code and doctest changes separate.
• Documentation change?

Reproduction and verification:

import numpy as np
from linear_algebra.src.power_iteration import power_iteration

matrix = np.diag([-4.0, 2.0])
value, vector = power_iteration(matrix, np.ones(2))
assert np.isclose(value, -4.0, rtol=0, atol=1e-9)
assert np.linalg.norm(matrix @ vector - value * vector) < 1e-5

Validation:

• All 5 existing doctests pass, including the real and Hermitian checks in test_power_iteration().
• Six additional local regression cases (three negative-dominant real/Hermitian matrices, each with flat and column vectors) fail before the change and pass afterward against numpy.linalg.eigvalsh and eigenvector residuals.
• ruff check and ruff format --check pass for the changed file.
• The repository's full pre-commit run --all-files check passes.
• Local numerical checks used Python 3.12 and NumPy 2.3.5. The full Python 3.15 repository test suite was not run locally.

Upstream CI: Ruff, type checking, documentation build, and pre-commit checks passed. The test run reports 3,428 passed and one failure in the untouched web_programming/crypto_price_tracker.py doctest: CoinGecko's live API returned HTTP 403. The power-iteration tests passed in that run.

Checklist

• I have read CONTRIBUTING.md.
• This pull request is all my own work -- I have not plagiarized.
• I know that pull requests will not be merged if they fail the automated tests.
• This PR only changes one algorithm file.
• All filenames are in all lowercase characters with no spaces or dashes.
• All functions and variable names follow Python naming conventions.
• All function parameters and return values retain their Python type hints.
• All existing doctests pass.
• No new algorithm or Python file is introduced.