x86/alternative: Rewrite optimize_nops() some

Address two issues:

 - it no longer hard requires single byte NOP runs - now it accepts any
   NOP and NOPL encoded instruction (but not the more complicated 32bit
   NOPs).

 - it writes a single 'instruction' replacement.

Specifically, ORC unwinder relies on the tail NOP of an alternative to
be a single instruction. In particular, it relies on the inner bytes not
being executed.

Once the max supported NOP length has been reached (currently 8, could easily
be extended to 11 on x86_64), switch to JMP.d8 and INT3 padding to
achieve the same result.

Objtool uses this guarantee in the analysis of alternative/overlapping
CFI state for the ORC unwinder data. Every instruction edge gets a CFI
state and the more instructions the larger the chance of conflicts.

  [ bp:
  - Add a comment over add_nop() to explain why it does it this way
  - Make add_nops() PARAVIRT only as it is used solely there now ]

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20230208171431.373412974@infradead.org