Replace get_this / update_this probing

[wlewNV]

Fixes #758

[Work-in-Progress]
PLACEHOLDER DESCRIPITON

[wlewNV]

Here is a small benchmark I ran with the change:

import timeit, statistics
from slangpy.slangpy import unpack_arg  # C++ version

# Test objects - typical args that don't have get_this
test_args = [42, 3.14, 'hello', [1, 2, 3], {'a': 1, 'b': 2}, (1, 2)]

def bench():
    for arg in test_args:
        unpack_arg(arg)

for _ in range(1000): bench()  # warmup
times = [(timeit.timeit(bench, number=10000) / 10000) * 1e9 for _ in range(5000)]
print(f'Mean: {statistics.mean(times):.2f} ns, Stdev: {statistics.stdev(times):.2f} ns')

For 5000 iterations × 10k calls each, 6 args per call, I get the following results:

• main (uncached nb::hasattr): 2136.67 ns ± 20.69 ns
• perf_win_2 (cached type lookup): 578.03 ns ± 10.30 ns

We are seeing a ~73% improvement (~1559 ns saved per call)

[ccummingsNV]

Hi - that's a good start, but the optimization we have in mind is a bit more in depth.

The idea here is to eliminate the need for get_this and update_this altogether, by making a base type from which a user can inherit. we could have the get_this/update_this checks right at the end as a fallback, but we want to avoid ever paying this cost really - any microseconds is too many microseconds

In your test case, I'd have expected you to see a saving in the case of a dictionary. The other types should bail out before hitting the get_this/update_this checks.