Address review feedback: add half-floats case to test_numbers(), simplify NEWS wording

Per skirpichev's review on GH-154569:
- Add a float16 (LE/BE) case to ArrayReconstructorTest.test_numbers()
  for basic functional coverage alongside the other numeric types,
  complementing the existing dedicated test_float16_endianness (which
  he confirmed is fine to keep for deterministically forcing the
  buggy slow path regardless of test-machine endianness).
- Simplify the NEWS entry wording per his suggested rewrite.

Author: PhysicistJohn

Lib/test/test_array.py

@@ -209,6 +209,10 @@ def test_numbers(self):
              [-1<<63, (1<<63)-1, 0]),
             (['l'], SIGNED_INT64_BE, '>qqq',
              [-1<<63, (1<<63)-1, 0]),
+            (['e'], IEEE_754_FLOAT16_LE, '<eeee',
+             [1.0, float('inf'), float('-inf'), -0.0]),
+            (['e'], IEEE_754_FLOAT16_BE, '>eeee',
+             [1.0, float('inf'), float('-inf'), -0.0]),
             (['f'], IEEE_754_FLOAT_LE, '<ffff',
              [16711938.0, float('inf'), float('-inf'), -0.0]),
             (['f'], IEEE_754_FLOAT_BE, '>ffff',

Misc/NEWS.d/next/Library/2026-07-23-21-00-00.gh-issue-154568.float16-reconstructor.rst

@@ -1,8 +1,5 @@
 Fix ``array._array_reconstructor`` (used internally for unpickling
-:class:`array.array`) ignoring the requested byte order for the
-``IEEE_754_FLOAT16_LE``/``IEEE_754_FLOAT16_BE`` machine format codes: it
-compared against the 32-bit float constant instead of the float16 one, so
-it always decoded as big-endian regardless of what was requested. This
-could silently produce wrong values when unpickling an ``'e'``-typecode
-(half precision float) array on a machine with different native
-endianness than the one that pickled it.
+:class:`array.array`) ignoring the requested byte order and always
+decoded as big-endian.  This could silently produce wrong values when
+unpickling a half precision float array on a machine with different
+native endianness than the one that pickled it.