Fix ROLZ inverse overread and transform test crashes

[BICHENG]

Fixes #37.

This fixes the crash from #37. I hit it through 7zFM + 7-Zip-zstd, then reduced it to kanzi-cpp.

Changes

• ROLZCodec1::inverse() now passes only the encoded bytes left after srcIdx to the bitstream reader. Small ROLZ blocks no longer get a 64 KiB input window when fewer bytes are available.
• TestTransforms.cpp moves the 1 MiB temporary buffer from the stack to heap storage.
• Transform tests release the current transform object on early returns.

Checks

On upstream commit f5839572, an MSVC x64 build can reproduce the crash with:

testTransforms -type=ROLZ -noperf
testTransforms -type=SRT -noperf
testTransforms -type=RANK -noperf
testTransforms -type=all -noperf

On this branch, the same commands pass.

The CI check does this in two steps:

• first it builds f5839572 with MSVC x64 and expects at least one of those transform commands to fail;
• then it builds this branch and expects all of them to pass.

That ties the CI check directly to #37.

It also compresses and decompresses generated random, structured, and mixed 4 MiB files with Kanzi CLI levels 1 to 9. SHA-256 hashes match after decompression.

Covered platforms: Ubuntu gcc, Ubuntu clang, macOS clang, MSVC x64, MSVC x86.

CI run:
https://github.com/BICHENG/kanzi-cpp/actions/runs/27264743969

[BICHENG]

Cleaned up the PR a bit
The ROLZ crash led to the SliceArray fix here. The SRT/RANK/all crashes look separate after another check, so I am leaving the test stack issue out of this PR.

[flanglet]

Hey, thanks for the changes.
I would have merged the previous PR but I will not merge this one. It is way too big and touches too many files (Lines changed: 583 additions & 191 deletions). There is too much code churn even for cosmetic reasons. I want to fix the actual transform issues where they happen.
What are the SRT/RANK/all crashes? Do they happen in the test code only?

[sonarqubecloud]

Quality Gate passed

Issues
3 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[BICHENG]

Yeah... I made this PR too broad.
Pushed it back to the smaller version.

The SRT/RANK/all crashes look like the test program using too much stack on Windows, so I kept that as a test-side fix.