Add concrete keyword to descendents

[maximlt]

Somewhat related to #651, the idea being to steer users towards descendents instead of concrete_descendents (to deprecate it ultimately but it'll take a little while as it's used in a lot of holoviz code bases).

[codecov]

Codecov Report

Attention: Patch coverage is 87.50000% with 1 line in your changes missing coverage. Please review.

Project coverage is 87.26%. Comparing base (274632a) to head (a313999).
Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
param/_utils.py	87.50%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1027      +/-   ##
==========================================
- Coverage   87.27%   87.26%   -0.02%     
==========================================
  Files           9        9              
  Lines        4936     4939       +3     
==========================================
+ Hits         4308     4310       +2     
- Misses        628      629       +1     

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

[jlstevens]

The concept here makes sense to me.

Discussing this with Maxime, we decided there are two paths we can take:

1. Have concrete_descendents warn the user when a key is clobbered, directing the user to using descendents with concrete=True
2. Work the usual deprecation cycle where we eventually get rid of concrete_descendents entirely (i.e. start warning users of .concrete_descendents to use descendents with the keyword instead)

These things are orthogonal and I think the first suggestion is a good idea regardless. I do see the argument for deprecating concrete_descendents (i.e. cleaning up the API a bit) , especially given that most usage seems to call .values() to get a list-like iterator anyway.

This PR is necessary whether we decide to deprecate concrete_descendents or not so I'll go ahead and approve it now.

[jlstevens]

Looks good to me!

[hoxbro]

(I haven't looked where this is used; if it is not in many places, ignore me).

Is there a penalty related to this check? E.g. if this is in a hotpath will this add significant overhead?

[maximlt]

Timed it and it's 2-3x slower than the try/except that follows. That's also not a particularly hot path I think.