Support symbolic integer scalars

@kshitij12345 is already working on this.

Since Dec 11, 2025.

## 🚀 Feature

Thunder JIT needs to support symbolic integer scalar values similar to PyTorch's Dynamo, which uses `SymInt` type annotations. This is critical for handling dynamic shapes and values that are only known at runtime.

## Current Behavior
Thunder likely treats all integer inputs as concrete values, causing recompilation for each new integer value.

## Expected Behavior
Thunder should support symbolic integer scalars that can represent runtime values, enabling:
- Symbolic type annotations in traces (e.g., `i0: SymbolicInt`)
- Tracking symbolic values through computation
- Proper code generation for dynamic values

## Motivation
- Pattern appears in KV-cache management for HF transformers

## Minimal Reproduction Case
```py
import torch
import thunder

@thunder.jit
def update_cumulative_length(cumulative_length: int, seq_len: int = 1) -> int:
"""cumulative_length should be treated as symbolic."""
return cumulative_length + seq_len

cumulative_length = 1024
new_length = update_cumulative_length(cumulative_length)
assert new_length == 1025
# With torch.compile(dynamic=True), cumulative_length becomes Sym(variable_name)
# Thunder should support similar symbolic tracking without concretizing values
print(update_cumulative_length._lc_cs.last_epilogue_traces[-1])
# def epilogue():
# return 1025
```

## Success Criteria
- [ ] Thunder can accept and track symbolic integer scalars
- [ ] Symbolic values maintain their symbolic nature through function calls
- [ ] No recompilation is happening on new values
- [ ] Proper type annotations for symbolic scalars
- [ ] Documentation on how to declare symbolic scalars