Fix precision loss in cagnoli for edges crossing the antimeridian

[thevar1able]

For an edge whose endpoints lie on opposite sides of ±180° the raw longitude difference in cagnoli is close to ±2π, and the true tiny step between the vertices is lost to the ulp of 2π before sin and cos are applied. Nothing cancels the error, so cells straddling the antimeridian at resolutions 12 and above come out too small: 0.07% at res 13, 0.4% at res 14, 2.5% at res 15 (cellAreaRads2 of the straddling cell vs. its neighbour).

The fix shifts both longitudes to the same side of the antimeridian before subtracting, so the difference is between two nearby values. sin and cos are 2π-periodic, so the result is unchanged mathematically. Global area sums at res 0–3 still equal 4π to within 5e-13. Of the cells used by the ClickHouse area tests only two antimeridian cells at res 0/1 change, in the 15th digit.

Adds a cell_area_antimeridian test with three straddling/neighbour cell pairs; it fails on the unpatched code and passes with the fix. testH3CellArea and testH3CellAreaExhaustive pass.

Regression from v4.5.0 (uber#1101), not present in v4.4.1 which used a different formula. To be reported upstream.

Closes: ClickHouse/ClickHouse#124447
Related: uber#1101
Related: uber#1108

🤖 Generated with Claude Code

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

Konstantin Bogdanov seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}