When your threat model is "someone on the network intercepting messages", it doesn't matter if you control the hardware. When your threat model is "someone owns my device", it still doesn't matter if you control the hardware, because, in that scenario, Apple is the bad actor you're trying to protect against.

There's no scenario where a third party has compromised your phone without Apple's collaboration, which is the only scenario where the secure enclave would maybe protect you (and even then, the bad actor would just read your messages off the screen or memory directly).

In principle you can do the same thing of having keys in a secure enclave that can only be accessed if the bootloader and OS were signed with an appropriate key and not revoked. In practice there would certainly be a larger attack surface because you've now got n different hardware secure enclaves, n different bootloaders, and n different OS implementations, and a flaw in any one of them is potentially all that an attacker needs. Would you allow apple to apply a high standard and e.g. blacklist manufacturers who repeatedly had holes in their implementation? Would you trust Huawei's implementation to not have a hidden backdoor accessible only to the Chinese state and not discoverable otherwise? (Do you trust Apple's implementation to not have the same for the US/Israel?)

Safety wise, Signal (or PGP email) on GrapheneOS is probably as good as it gets since it's all audited free open source code.