Fix inner product distance code to vectorize on arm (affects all arch)

[pmiloslavsky]

Summary

Fixed inner product code to vectorize by removing inner loop use of void.

Changed example_search to also demonstrate use of IP space with normalization and print index build timing.

Validation

Disassembled example_search InnerProductDistance with gdb and checked for fmla instruction on ARM, Ubuntu 24 ARM.

gdb -batch -ex 'disassemble/rs _ZN7hnswlibL20InnerProductDistanceEPKvS1_S1_' ./example_search

Ran
python -m unittest discover --start-directory tests/python --pattern "bindings_test*.py"

Before fix (but with example_search.cpp copied over):

   0x0000000000002b34 <+52>:	1f000441	fmadd	s1, s2, s0, s1
...
pmilosla@iscinternal.com@qdub24arm1:/nethome/pmilosla/projects/hnswlib/testbed_linux_arm/hnswlib_clean/hnswlib$ build/example_search 
Running l2 test...
l2 index build time: 1.03063 seconds (1030.63 ms)
l2 Recall: 1
l2 Recall of deserialized index: 1

Running ip_normalized test...
ip_normalized index build time: 1.18953 seconds (1189.53 ms)
ip_normalized Recall: 1
ip_normalized Recall of deserialized index: 1

After fix (note decreased time for ip):

   0x0000000000002b30 <+48>:	4e21cc40	fmla	v0.4s, v2.4s, v1.4s
...
pmilosla@iscinternal.com@qdub24arm1:/nethome/pmilosla/projects/hnswlib/testbed_linux_arm/hnswlib_fixed/hnswlib$ build/example_search 
Running l2 test...
l2 index build time: 1.02962 seconds (1029.62 ms)
l2 Recall: 1
l2 Recall of deserialized index: 1

Running ip_normalized test...
ip_normalized index build time: 0.999375 seconds (999.375 ms)
ip_normalized Recall: 1
ip_normalized Recall of deserialized index: 1

[pmiloslavsky]

@kiplingw