Normalize CoDICE `epoch_delta_minus` and `epoch_delta_plus` across the
five L2 products that currently emit them:

- `imap_codice_l2_hi-omni`
- `imap_codice_l2_hi-sectored`
- `imap_codice_l2_lo-sw-species`
- `imap_codice_l2_hi-direct-events`
- `imap_codice_l2_lo-direct-events`

Before this change, the products were split across three inconsistent
states:

- `hi-omni` already wrote `epoch_delta_*` as `CDF_INT8`, but the final
  output `epoch` variable did not retain `DELTA_MINUS_VAR` /
  `DELTA_PLUS_VAR`
- `lo-sw-species` still wrote `epoch_delta_*` as floating-point values
  with float-style metadata, so SPDF accepted the file even though it
  did not follow the intended TT2000-style integer-duration convention
- direct-events still wrote `epoch_delta_*` as `CDF_DOUBLE` while the L2
  attrs were already trying to describe them as integer support
  variables, which caused SPDF validator failures on datatype, format,
  and fill semantics

Fix the shared CoDICE timing source so delta durations are computed as
exact integer nanoseconds.

Specifically:
- update `get_codice_epoch_time()` to compute `delta_times` with integer
  arithmetic end-to-end
- keep the center-time calculation in floating-point seconds, but feed it
  integer `delta_times`
- avoid float-to-int conversion entirely, so the written delta variables
  naturally serialize as `CDF_INT8`

Also normalize the CoDICE `epoch_delta_*` metadata to one shared
integer-duration convention:
- `FILLVAL = -9223372036854775808`
- `FORMAT = I19`
- `VALIDMIN = 0`
- `VALIDMAX = 128000000000`

The chosen `VALIDMAX` corresponds to the current CoDICE timing model:
- `delta = num_spins * spin_period / 2`
- `spin_period VALIDMAX = 16 s`
- `num_spins max = 16`
- worst-case delta = `128 s = 128000000000 ns`

Apply that metadata in:
- the shared CoDICE L1A attrs
- the explicit L2 `epoch_delta_*` overrides for:
  - hi-omni
  - hi-sectored
  - hi-direct-events
  - lo-direct-events

Finally, attach the delta-variable links to the final emitted `epoch`
variable in all five target L2 products:
- `DELTA_MINUS_VAR = "epoch_delta_minus"`
- `DELTA_PLUS_VAR = "epoch_delta_plus"`

This is done on the final L2 dataset rather than on intermediate source
datasets so the links survive writeout consistently.

Extend the written-CDF regression coverage to verify:
- representative CoDICE L1A products now write `epoch_delta_*` as
  `CDF_INT8`
- `hi-omni` and `hi-sectored` write `CDF_INT8` deltas with the normalized
  metadata and carry the `epoch` delta links
- `lo-sw-species`, `hi-direct-events`, and `lo-direct-events` do the
  same

Validated locally with:
- `pytest -q imap_processing/tests/codice/test_codice_l1a.py`
- `pytest -q imap_processing/tests/codice/test_codice_hi_l2.py`
- `pytest -q imap_processing/tests/codice/test_codice_l2.py -k 'sw_species or lo_de or hi_de'`
- `pre-commit` on the touched CoDICE files

Move the CoDICE `epoch` delta-link metadata out of `codice_l2.py` and
into the product variable-attrs YAMLs, as suggested in review.

This keeps the `DELTA_MINUS_VAR` / `DELTA_PLUS_VAR` links declared in the
same metadata layer as the rest of the emitted CDF attributes, instead
of mutating the final `epoch` attrs in code. The affected products are:

- `imap_codice_l2_hi-omni`
- `imap_codice_l2_hi-sectored`
- `imap_codice_l2_lo-sw-species`
- `imap_codice_l2_hi-direct-events`
- `imap_codice_l2_lo-direct-events`

Also remove the now-unneeded in-code helper that attached those attrs to
the final dataset.

One follow-up change is needed in the `lo-sw-species` processing path:
after the intensity calculation, refresh the `epoch` attrs from the L2
attr manager so the YAML-defined delta links are carried into the final
written file. The other CoDICE L2 paths were already reapplying `epoch`
attrs before writeout.

Finally, add a short test comment explaining the provenance of
`EXPECTED_EPOCH_DELTA_VALIDMAX = 128000000000`, since that bound is
derived from the current CoDICE timing model rather than from the raw
`int64` storage limit.

Validated locally with:
- `pytest -q imap_processing/tests/codice/test_codice_l1a.py`
- `pytest -q imap_processing/tests/codice/test_codice_hi_l2.py`
- `pytest -q imap_processing/tests/codice/test_codice_l2.py -k 'sw_species or lo_de or hi_de'`
- `pre-commit run --files ...`