Fix SPD on Gimlet and Cosmo (#617)

This is a fix for humility#596, humility#595, and humility#568.

In hubris#2094, we moved the SPD data to a parameterized `struct SpdData<const
DIMM_COUNT: usize, const DATA_SIZE: usize>`. This is because Cosmo has a
different number of DIMMS (16 → 12), and their SPD data are a different size
(512 → 1024 bytes).

Unfortunately, this broke Humility, which was expecting an array named
`spd_data` and instead found a `struct`.

Cosmo is _also_ broken because DDR5 has a different SPD layout than DDR4; even
if we could read the `struct`, we wouldn't know how to parse it.

This PR fixes both things, flexibly decoding the `SpdData` and printing both
DDR4 and DDR5 SPDs.  (This required updating the `spd` crate with DDR5 info) 

To avoid a giant diff, this doesn't add new dumps for the test suite, but I'm
planning to add them in a subsequent PR.

Here's an example of Cosmo output:
```
     Running `target/debug/humility -d /Users/mjk/Desktop/hubris.core.cosmo spd`
humility: attached to dump
ADDR MANUFACTURER              PART                           WEEK YEAR
   0 Micron Technology         MTC40F2047S1RC64BB1 USFF          2 2025
   1 Micron Technology         MTC40F2047S1RC64BB1 USFF          2 2025
   2 Micron Technology         MTC40F2047S1RC64BB1 USFF          2 2025
   3 Micron Technology         MTC40F2047S1RC64BB1 USFF          2 2025
   4 Micron Technology         MTC40F2047S1RC64BB1 USFF          2 2025
   5 Micron Technology         MTC40F2047S1RC64BB1 USFF          2 2025
   6 Micron Technology         MTC40F2047S1RC64BB1 USFF          2 2025
   7 Micron Technology         MTC40F2047S1RC64BB1 USFF          2 2025
   8 Micron Technology         MTC40F2047S1RC64BB1 USFF          2 2025
   9 Micron Technology         MTC40F2047S1RC64BB1 USFF          2 2025
  10 Micron Technology         MTC40F2047S1RC64BB1 USFF          2 2025
  11 Micron Technology         MTC40F2047S1RC64BB1 USFF          2 2025
```

Author: mkeeter