[cppyy] Add converters and low-level views for fixed width integers

[aaronj0]

fixes: #17841

The support for the fixed-width types int16_t and int32_t indeed diverges from upstream cppyy and uses unsigned long which triggers the incorrect numpy view. Despite not having converters for the same, upstream relies on patches (or older behaviour) along the Cppyy::ResolveName -> TClassEdit::ResolveTypedef code path (see ResolveTypeDefImpl) that resolved fixed width integer types to their standard type counterparts (int16_t[][] to short[][] and int32_t[][] to int[][]) when creating converters. Trying to follow the same code branch in ROOT can have a large blast radius in the type system. Two non-intrusive ways to fix this issue are either aliasing to existing short and int converters, or adding the converters and low-level views for both types. This PR performs the latter

[github-actions]

Test Results

    19 files      19 suites   4d 16h 39m 46s ⏱️
 2 740 tests  2 740 ✅ 0 💤 0 ❌
50 623 runs  50 623 ✅ 0 💤 0 ❌

Results for commit 7b1108b.

♻️ This comment has been updated with latest results.

[guitargeek]

Thank you very much for implementing better support for fixed-width integers! Looks good to me, but please also add a test 🙂

[guitargeek]

Oh and don't forget to fix also the compiler warnings, otherwise the debian125 build that treats warnings as errors because of dev=ON will fail

[aaronj0]

The following issue: #9809 causes failures on windows x86, where the arrays assigned on the C++ side itself are indexed/read wrongly:

root [0] int test[]={2,4,6,8,10};
root [1] test[4]
(int) 2
root [2] test[4]=20;
root [3] test[4]
(int) 20
root [4] test[0]
(int) 20
root [5] test
(int [5]) { 20, 4, 6, 8, 10 }
root [6]

Hence validating those buffers fail and the test will have to be disabled on win x86.

[aaronj0]

Looks like the failures on win 64 are unrelated: roottest-root-io-stdpair