seagate: clamp fw-activate-history entry count, bound JSON string copies

[prabhakarpujeri]

Problem

The Seagate 0xC2 vendor log page is device-supplied, and the
fw-activate-history display paths trusted two fields in it:

1. numValidFwActHisEnt (a device u32) was used directly as the loop
   bound against the compiled-in fwActHisEnt[20] table — a count >
   20 walks past the table in both the plain and JSON views.
2. The JSON view copied previousFW / newFW with
   sprintf(char[8], "%s", …) from fixed-width u8[8] fields that carry
   no NUL guarantee — a malformed page gives a device-controlled stack
   overflow (the plain view copies by size into 9-byte buffers and is
   safe).

Fix

• Clamp numValidFwActHisEnt to the table size immediately after the
  single log read, so both display paths can only walk real entries.
• Make the JSON view copy the fixed-width strings by size into 9-byte
  zeroed buffers, exactly mirroring the plain-print path construction.

The clamp mutates only the local stack struct and only on the success
path; nothing prints the raw count afterwards, so the display simply
tops out at the table size.

Validation

• Full meson build clean (nvme.link, all plugins).
• ASan/UBSan harness replicating the exact old and new expressions with
  adversarial fixtures (numValidFwActHisEnt = 0xffffffff, 8-byte FW
  fields without NULs): pre-fix path reports stack-buffer-overflow in
  old_json_path; fixed path passes clamp-max-20, pass-through on
  small counts, and 8-byte NUL-termination checks (4/4).

[igaw]

Thanks!