Optimize interruptible key agreement dispatch through ECDH

[gilles-peskine-arm]

PSA interruptible key agreement dispatches has a long dispatch chain:

• interface in psa_crypto.c
• driver dispatch
• built-in implementation in psa_crypto_ecp.c
• Legacy ECDH in ecdh.c
• ECC arithmetic in ecp.c

The dispatch through ECDH has known inefficiencies where we're copying data around, but don't need to. The goal of this task is, at a minimum, to optimize this. Furthermore, look for other inefficiencies in the ECDH indirection and analyze how we might optimize those. The answer might be to eliminate ECDH altogether, or to change the ECDH interface. Note that starting in TF-PSA-Crypto 1.0, ecdh.h is an unstable or internal interface and we can change it at any time.