Preserve aligned Series weights and isolate reset-index weights

[juaristi22]

MicroSeries binary operations now retain weights with their row labels when pandas aligns operands. For values [10, 20], labels ['b', 'a'], and weights [1, 9], adding values [2, 1] at labels ['a', 'b'] now produces the weighted total 22 * 9 + 11 * 1 = 209; the operators previously returned 121 by reapplying weights positionally. Named arithmetic and comparison methods consistently use the calling MicroSeries' weights across pandas 2 and 3. Explicit reverse dispatch preserves weighted results when a plain pandas Series appears on the left.

MicroDataFrame.reset_index now copies weight storage on pandas 2, so changing a result's weights cannot change its source's weighted statistics. Regression tests cover both mutation directions, drop and in-place variants, and partial MultiIndex resets. Existing safeguards continue to reject unknown or ambiguous row weights.

Validation:

• Full suite: 543 tests passed on pandas 2.3.1 and pandas 3.0.5.
• Independent verification: 53 focused checks passed in each environment.
• After formatting: 157 binary-operation tests passed in each environment; executable syntax unchanged.
• Combined with Estimate variance from replicate weights #320 at commit 058f8db: 593 full-suite tests and 30 interaction tests passed in each environment.
• make lint and the towncrier draft passed.

Related #170. A warning for mismatched input weights remains a separate enhancement.