objdump 2.46.1 prints {bad} for VCVTPS2PH encoding generated by NASM

I am developing my own x86-64 assembler and recently implemented EVEX
`{sae}` handling.

I encountered a potentially incorrect disassembly of `VCVTPS2PH`.

The source I tested is:

```asm
section .text
global _start

_start:
vcvtps2ph [rax+rcx]{k2}, zmm2, {sae}, 10
```

NASM 3.02 accepts this instruction and produces the following bytes:

```text
62 f3 7d 1a 1d 14 08 0a
```

My assembler produces exactly the same bytes.

However, GNU objdump 2.46.1 disassembles them as:

```text
vcvtps2ph [rax+rcx*1]{bad}{k2},xmm2,0xa
```

I also tested this with an older objdump (2.40), which produces the same
`{bad}` result.

Could you please clarify whether:

1. this encoding is architecturally invalid,
2. NASM is accepting an invalid encoding, or
3. objdump's disassembler does not correctly decode this particular EVEX
form?

The relevant bytes are:

```text
62 f3 7d 1a 1d 14 08 0a
```

The Intel SDM description of `VCVTPS2PH` does not make the handling of
`{sae}` in this particular form sufficiently clear to me.

Versions:


NASM: 3.02
objdump: 2.46.1
OS: Debian 12 x86-64 (bookworm)


Thank you for any clarification.

Best regards,
Ammar