Add cftime support

[ahuang11]

Closes #144

After attempting this, I realized it did not feel like a good first issue~ Note I used Claude Opus to plan and write this.

Two tiers based on calendar type:

1. Gregorian-like (standard, noleap, all_leap) → pa.timestamp('us'). String SQL filters work naturally (WHERE time > '1980-01-01'). Microsecond resolution covers ±292k years, avoiding the nanosecond 1678–2262 overflow.
2. Non-Gregorian (360_day, julian) → pa.int64() with xarray:units/xarray:calendar metadata on the Arrow field. Lossless — a 360-day "Feb 30th" isn't forced into Gregorian. Auto-registered cftime() UDF provides ergonomic filtering: WHERE time > cftime('2000-07-01').

Now this runs:

import xarray as xr
from metpy.cbook import get_test_data
from xarray_sql import XarrayContext

ds = xr.tutorial.open_dataset("rasm")
context = XarrayContext().from_dataset("rasm", ds, chunks={"time": 2})
context.sql("SELECT * FROM rasm LIMIT 1000")

However as you can see, there are a lot of NaNs; to be fixed in a separate PR in #151

[alxmrs]

After attempting this, I realized it did not feel like a good first issue

Ok, full confession: I may be bad at scoping what a good first issue is or not. It's also hard to know what's involved before you've implemented the thing. I appreciate this feedback.

[alxmrs]

Thanks for this PR! Here is my initial round of feedback. Mostly questions, some suggestions.

[ahuang11]

Thanks for the review; a bit busy this early week, but will definitely get to this later. I'm also planning to first make a docs infra PR first so this can build on that docs.