feat: PCI host routing for Card/Token instruments, expand MetaInfo

[282000ypk]

• Add PciHostFilter and DeviceOsHeaderFilter via InstrumentRequestFilter interface
• Route Card/Token instruments to cards.phonepe.com via PayContext filter chain
• Expand MetaInfo from udf1-5 to udf1-15 with @Size/@pattern constraints

[sk4x0r]

Review: 5 inline comments covering validation bypass, auth-retry semantics, repetitive validation, module-level constant placement, and serialization config duplication.

[sk4x0r]

Validation bypass via direct constructor

MetaInfo(udf11="invalid!!!$$$") succeeds silently — validation only runs inside build_meta_info(). Anyone constructing MetaInfo directly (or via deserialization with from_dict/from_json) skips all size and pattern checks.

Consider adding a __post_init__ hook that runs the same validation, or at minimum document prominently that build_meta_info() is the only safe entry point.

[sk4x0r]

Auth-retry path — method name is misleading

The new http_command parameter is wired correctly (the local command variable is used for the request). However, the method name _request_via_auth_refresh implies it retries after refreshing the token, but it actually re-raises the UnauthorizedAccess exception after calling force_refresh_token() (line 100). The caller never gets a retry with the refreshed token.

This is pre-existing behavior — not introduced by this PR — but now that http_command makes the routing caller-visible, it's worth flagging: if a retry is ever added, it must reuse command (not self._http_command).

[sk4x0r]

Suggestion: replace repetitive validation with a loop

15 explicit _validate_size / _validate_size_and_pattern calls are error-prone and tedious to extend when new fields are added. Consider:

fields = locals()
for i in range(1, 11):
    cls._validate_size(f"udf{i}", fields[f"udf{i}"], cls._FREE_MAX)
for i in range(11, 16):
    cls._validate_size_and_pattern(f"udf{i}", fields[f"udf{i}"], cls._RESTRICTED_MAX)

This keeps the same semantics with ~4 lines instead of 15.

[sk4x0r]

Nit: _PCI_INSTRUMENT_TYPES placement

This module-level constant sits between the import block and the class definition, which reads a bit awkwardly. Consider moving it inside CustomCheckoutClient as a class-level constant (e.g., _PCI_INSTRUMENT_TYPES = ...) or placing it right next to the pay() method where it's used, to improve locality.

[sk4x0r]

Suggestion: deduplicate exclude=lambda x: x is None

The same metadata=config(exclude=lambda x: x is None) is repeated on all 15 fields. A module-level constant would reduce noise:

_EXCLUDE_NONE = config(exclude=lambda x: x is None)

@dataclass
class MetaInfo:
    udf1: Optional[str] = field(default=None, metadata=_EXCLUDE_NONE)
    udf2: Optional[str] = field(default=None, metadata=_EXCLUDE_NONE)
    ...