Fix parallel load

[jlamypoirier]

✨ Description

Fix: #244

This should fix all the bugs described in #244:

• Parameter counter check now takes duplicates into account (fix tensor and pipeline-parallel)
• Avoid _set_implicit_default for sequence_tensor_parallel since it's an override and not an implicit default. (fix the bug reported for sequence data parallel, though it doesn't technically relates to the feature)

Also address other minor issues and failing tests, and a tentative fix for #249.

🔍 Type of change

Select all that apply:

• 🐛 Bug fix (non-breaking change that addresses a specific issue)
• 🚀 New feature (non-breaking change that adds functionality)
• ⚠️ Breaking change (a change that could affect existing functionality)
• 📈 Performance improvement/optimization (improves speed, memory usage, or efficiency)
• 🛠️ Code refactor (non-functional changes that improve code readability, structure, etc.)
• 📦 Dependency bump (updates dependencies, including Dockerfile or package changes)
• 📝 Documentation change (updates documentation, including new content or typo fixes)
• 🔧 Infrastructure/Build change (affects build process, CI/CD, or dependencies)