quantile/nanquantile fixes

[crusaderky]

This rabbit hole took me WAY longer than I anticipated.

quantile/nanquantile changes

• Fix and test weights parameter (close da.quantile with weights fails #12322)
• Test when quantiles are a NumPy or Dask array
• Add support for n-dimensional quantiles
• Test that dtype propagation matches numpy
• Raise if the user passes overwrite_input=True

numbagg changes

• Fix crash in nanquantile when numbagg is installed and quantiles are an array (as opposed to a list)
• Disable numbagg in nanquantile when a.dtype==np.float32 (produces incorrect output dtype)
• Document minimum supported version of numba and numbagg
• Drop untested support for numbagg 0.7.0 (impacts nanmedian)

[github-actions]

Unit Test Results

See test report for an extended history of previous test failures. This is useful for diagnosing flaky tests.

     25 files  ±    0       25 suites  ±0   6h 43m 28s ⏱️ + 8m 31s
 18 634 tests +  186   17 249 ✅ +  186   1 385 💤 ±    0  0 ❌ ±0 
378 863 runs  +4 092  328 081 ✅ +2 795  50 782 💤 +1 297  0 ❌ ±0 

Results for commit 14fea53. ± Comparison against base commit b32233c.

♻️ This comment has been updated with latest results.

[crusaderky]

@dask/maintenance I could use an extra set of eyes on this