Why does the bitcoin protocol enforces no dublicate inputs at consensus

[hmbdofficial]

Looking the the transaction structure and bitcoin protocol enforcement, one will just wonder how the bitcoin protocol enforce the no dublicate inputs at the consensus level, and also what the security  implications it may cause If a soft work is proposed to relax this in favor of allowing replacement of inputs within the same block similar to certain covenant proposal.
I also thought about this from a protocol design perspective, why was the witness structure (segwit) was implemented as soft work rather than a hard fork, and also what specific risk it would arisen if the witness Markle root had been replaced in the coinbase transaction instead of the block header extension?

[ABCbits]

Looking the the transaction structure and bitcoin protocol enforcement, one will just wonder how the bitcoin protocol enforce the no dublicate inputs at the consensus level

Can you elaborate more? Bitcoin node have UTXO set (sometimes called chainstate) to prevent same input spend more than once and check whether transaction actually use unspend input that actually exist.

why was the witness structure (segwit) was implemented as soft work rather than a hard fork

I recall the main reason is backward compatibility, which increase chance it would be supported/adopted.