#17482: Add matmul validation to prevent a combination of width sharded with block sharded inputs

[edwinleeTT]

Ticket

17482

Problem description

Performing matmul with a block sharded + a width sharded input resulted in a device hang

What's changed

Added validation to error when invalid combination of width + block sharded inputs are received.

Checklist

• All post commit CI passes
• https://github.com/tenstorrent/tt-metal/actions/runs/13335717521
• Blackhole Post commit CI passes (if applicable)
• Model regression CI passes (if applicable)
• Device performance regression CI passes (if applicable)
• (For models and ops writers) Full new models tests CI passes (if applicable)
• New/Existing tests provide coverage for changes