[smart_holder] smart_holder / master benchmarking 2025-02-16

[rwgk]

Description

NOT FOR MERGING

This PR simply removes smart_holder tests so that

• tests/CMakeLists.txt is identical to master

while the code under

• include/pybind11 is exactly as under the current smart_holder branch.

This was used to collect the measurements shown in the

• smart_holder / master benchmarking 2025-02-16 spreadsheet.

The results in a nutshell:

• .so sizes of extension modules are expected to be <3% larger (smart_holder / master).
• Compile and link times are expected to be <2% larger.
• Run times are expected to be practically identical (based on pytest execution times).

Note that these are not micro benchmarks, but assumed to reflect real-life situations.

For easy future reference:

• Raw logs zipped and unpacked

• State of master branch

• State of smart_holder branch

Suggested changelog entry: