BUG: concat DatetimeIndex freq depends on input order (GH#64253)

[gaurav0107]

• closes BUG: The order of items in pd.concat matters for the resulting .index.freq. #64253
• Tests added and passed
• All code checks passed
• Added an entry in the latest doc/source/whatsnew/v3.1.0.rst file

pd.concat(..., axis=1) over objects whose indexes share a non-inferable
frequency (e.g. CustomBusinessDay) produced a combined index whose .freq
depended on the order of the inputs: a pairwise union can drop the shared
frequency in an intermediate, gappy step, and because CustomBusinessDay
cannot be recovered by inferred_freq it was never restored even after the
final union made the index regular again.

The combination path now reattaches the shared frequency when every input shares
a single frequency and the combined index is exactly regular with respect to it
(verified via a date_range equality check), so the result no longer depends
on input order. Index values and dtype are never modified; genuinely gappy or
mixed-frequency unions are left with freq=None as before.