DxilValidation: TGSM size check should be per-entry in lib target

Description
In DxilValidation.cpp, a check for the total size of TGSM (Thread Group Shared Memory) used counts all groupshared global variables in the module. This is not valid for a library with multiple entry points, when each of which only uses a subset of the groupshared globals. The correct per-entry groupshared usage is collected into the RDAT part, but the independent calculation in the validator is left-over from enforcing the maximum on compute targets at the module level instead.

This mainly impacts Work Graphs (node shader) for now, since that's the only runtime-supported compute entry in a DXIL library.

Steps to Reproduce
Repro here:
https://godbolt.org/z/z88esxr1s

Individually, these two entries are valid, but together in a lib, they produce a validation error due to the summed groupshared memory sizes.

Actual Behavior

error: validation errors
<source>:35: error: Total Thread Group Shared Memory storage is 65536, exceeded 32768.
Validation failed.

Environment

• DXC version: 1.8.2403
• Host Operating System: any

Start in DxilValidation.cpp and reproduce the failure with the Godbolt example linked in the issue. Compare the validator's module-level TGSM calculation with the per-entry usage already collected in RDAT. Done means a library with individually valid entries no longer fails because their groupshared sizes are summed together.