Promote Python 3.15 to a blocking CI target

[derek73]

Closes #259 (step 2).

Python 3.15.0 final is out. actions/python-versions marks it stable: true, and master's CI has been testing the final release since it went live: the allow-prereleases spec resolves to ~3.15.0-0, which matched 3.15.0 once it shipped (run 38125388105 installed python-3.15.0-linux-24.04-x64).

Commit 1: promote 3.15 to blocking

• python-package.yml: drop continue-on-error for 3.15, so a 3.15 failure now fails the build. Drop allow-prereleases: true too, so a future "3.16" entry can't silently pick up alphas.
• pyproject.toml: add the Programming Language :: Python :: 3.15 classifier.
• CONTRIBUTING.md: the stated CI range was stale (it said 3.10–3.14). It now points at the workflow matrix instead of restating it.
• docs/release_log.rst: add an Additions bullet.

Commit 2: run ja-extra on 3.15
The ja extra's compiled dependencies had no cp315 wheels at the versions uv.lock pinned, so building them from sdist failed. This commit bumps the four that need it:

package	was	now
numpy	2.5.1	2.5.4
scipy	1.18.0	1.18.1
pandas	3.0.5	3.0.6
regex	2026.7.19	2026.9.29

lightgbm ships a py3-none wheel, so it didn't need one. The older numpy and scipy (2.4.6 and 1.17.1, which the lock keeps for Python 3.11) are unchanged.

Commit 3: ja-extra runs on 3.11 and 3.15 (review fix)
The lock splits numpy and scipy at 3.12, and ja-extra had only ever run the newer branch. It is now a matrix over the floor and the newest Python, so both branches are installed and tested. Its comment states the rule (a new top row can need a lock bump first) instead of a snapshot. Locally on 3.11, tests/v2/ passes with numpy 2.4.6 and scipy 1.17.1.

Tested locally on 3.15.0 (macOS arm64), with source builds forbidden for those packages: the extra installs from wheels, and tests/v2/ passes (14201 passed). That includes the 771 ja, segmenter and namedivider tests, none of them skipped. The Linux run in this PR's CI is the check that matters.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.02%. Comparing base (3b2b6f0) to head (789f0dd).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #642   +/-   ##
=======================================
  Coverage   99.02%   99.02%           
=======================================
  Files          46       46           
  Lines        4511     4512    +1     
=======================================
+ Hits         4467     4468    +1     
  Misses         44       44           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.