Make INTERNAL output models robust to server-side `columns` filtering

[MarketDataDev03]

Summary

Follow-up to #27 (which fixed #23). The columns parameter on
options.expirations is applied server-side, so the API can omit any
field from the response — not just updated. PR #27 fixed the specific
columns=["expirations"] case and made updated optional, but the underlying
class of bug remains for other column combinations.

This is a low-priority hardening issue: in practice the API almost certainly
always returns expirations as the core payload, so these paths are unlikely
to trigger today. Filing to track the root cause rather than leave it implicit.

Root cause

The INTERNAL output format builds the dataclass via output_model(**data).
Any field that is not present in a filtered response (because the user
excluded it via columns) will raise TypeError if it's a required field.

Gaps still present

1. columns=["updated"] (or any filter excluding expirations) crashes the
   INTERNAL path. OptionsExpirations.expirations is still a required field:

   @dataclass
   class OptionsExpirations:
       s: str
       expirations: list[datetime.datetime]      # required
       updated: datetime.datetime | None = None  # optional (fixed in #27)
   
   A response without expirations → TypeError → MarketDataClientErrorResult
   — the same failure class as #23, just a different column.

2. OptionsExpirationsHumanReadable got no equivalent treatment. Both
   Expirations and Date are required, and post_init calls
   format_timestamp(self.Date) unconditionally. A partial human-readable
   INTERNAL response missing Date would crash:

@DataClass
class OptionsExpirationsHumanReadable:
Expirations: list[datetime.datetime] # required
Date: datetime.datetime # required; format_timestamp called unconditionally

Proposed fix

Treat every server-side-filterable field as optional in both INTERNAL models
(default None + guarded conversion in post_init), consistent with how
updated was handled in #27. Decide explicitly whether expirations /
Expirations should be optional or whether an empty payload should be rejected
with a clearer error — a response with no expirations data is arguably
meaningless, so a deliberate validation error may be preferable to silent None.

Acceptance criteria

• INTERNAL path does not raise TypeError for any valid columns filter
  combination on options.expirations.
• OptionsExpirationsHumanReadable handles partial responses consistently
  with OptionsExpirations.
• Regression tests covering columns=["updated"] and a partial
  human-readable response.

References

• Fixed by / follow-up to FIX/options.expirations returns an empty DataFrame when filtering columns #27
• Original bug: options.expirations returns an empty DataFrame when columns=["expirations"] is passed #23