fix(device-table): reject oversized entry counts

[AbdulRehman77777]

Summary

Reject device tables whose declared memory-region or peripheral counts exceed their fixed array capacities.

Problem

eos_device_table_validate() previously checked the table magic, version, and CRC, but did not validate mem_region_count or periph_count.

A malformed device table could therefore have a valid CRC while declaring more entries than its fixed arrays can store.

Code that trusts a successful validation result could then iterate past the bounds of the memory-region or peripheral arrays.

Solution

Add structural validation for both count fields before accepting the table:

• Reject mem_region_count > EOS_MAX_MEM_REGIONS
• Reject periph_count > EOS_MAX_PERIPHERALS
• Return the existing EOS_ERR_INVALID error for invalid structural metadata

Valid device tables and the public API remain unchanged.

Testing

• Added regression coverage for oversized memory-region counts.
• Added regression coverage for oversized peripheral counts.
• Confirmed the regression failed before the production fix.
• test_device_table: 8/8 passed after the fix.
• ASan/UBSan targeted tests: 8/8 passed.
• Unaffected CTest suite: 13/13 passed.
• Full CTest: 13/14 due to the existing test_recovery link failure on upstream.
• Config generation passed using the repository CI fixture.
• git diff --check passed.
• Independent review completed with no findings.

Impact / Limitations

This is a focused device-table validation fix with no API changes or new dependencies.

The existing unrelated test_recovery link failure and pre-existing compiler warnings are not addressed by this PR.