Fix weekly return aggregation on pandas >= 2.0 (DatetimeIndex.week removed)

[nyxst4ck]

Problem

utils.aggregate_returns() uses DatetimeIndex.week for the weekly grouping paths ("week", "eow", and "W"). That accessor was deprecated in pandas 1.1 and removed in pandas 2.0, so any weekly aggregation raises:

AttributeError: 'DatetimeIndex' object has no attribute 'week'

Since the project targets pandas >= 1.5 (and Python 3.10–3.13), this breaks for the large majority of users who are on pandas 2.x / 3.x.

The break also surfaces through the public stats API, which forwards aggregate="W" down to aggregate_returns:

import numpy as np, pandas as pd
import quantstats as qs

r = pd.Series(np.random.randn(60) * 0.01,
              index=pd.date_range("2020-01-01", periods=60, freq="D"))

qs.stats.best(r, aggregate="W")   # AttributeError on pandas >= 2.0

best, worst, win_rate, avg_return, avg_win, avg_loss, consecutive_wins and consecutive_losses are all affected.

Fix

Replace index.week with index.isocalendar().week. The old .week/.weekofyear attributes were aliases for the ISO week number, which is exactly what isocalendar().week returns, so this restores the original behavior without changing results. Pairing with index.year (as before) is preserved for the end-of-week grouping.

Tests

Added regression tests in tests/test_utils.py covering weekly aggregation via both aggregate_returns ("W" and "week") and the stats public API (best/worst with aggregate="W"), including a value check against an explicit isocalendar-based compounding reference. The new tests fail on main (AttributeError) and pass with this change; the full existing suite continues to pass.

[nyxst4ck]

Marked ready for review. I do not see any checks reported on this branch yet.