PERF: cascade Series.dt.components for ArrowDtype durations

[aboderinsamuel]

• closes #xxxx (Replace xxxx with the GitHub issue number)
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions.
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file.
• I have reviewed and followed all the contribution guidelines.
• If I used AI to develop this pull request, I prompted it to follow AGENTS.md.

Speeds up Series.dt.components for ArrowDtype durations with two changes:

1. Sub-day fields are peeled off a running remainder from the coarsest unit down, so each is a single division with no modulo (PyArrow has no modulo kernel, so each % was emulated with divide/multiply/subtract).
2. The day floor-division is cached once and shared by _dt_days and _dt_day_remainder, instead of being recomputed on every .dt.components call.

No behavior change, covered by the existing test_arrow duration-component tests.

Benchmark of TimedeltaComponents.time_components (duration[ns][pyarrow], best of 60, before = main / after = branch): Series.dt.components 34.5 ms -> 18.0 ms (~1.9x) at N=100k all-positive, and 320 ms -> 173 ms (~1.85x) at N=1M mixed-sign. Single-field accessors unchanged.

xref #63283, #63470

[jbrockmendel]

Can you check perf on ser.dt.days? I suspect this is non-trivially slower

[jbrockmendel]

Closing as stale. Can reopen when you're ready to address comments and continue.