[AI-FSSDK] [FSSDK-12760] add localHoldouts to datafile for backward compatibility

[jaeopt]

Summary

Adds support for the new top-level localHoldouts datafile section so this SDK can be served the same backward-compatible datafile shape that Python (#517) and Ruby (#400) already consume.

Section membership — not includedRules — is now the sole signal for holdout scope:

• holdouts → ALL entries are global. Any stray includedRules on these entries is stripped at parse time (with a warning log).
• localHoldouts → ALL entries are local. Entries without a non-empty includedRules list are invalid and skipped with an error log (no fallback to global application).

Old datafiles that emit only holdouts continue to parse unchanged.

Why this is safe in production

1. No public API changes — HoldoutConfig is internal (struct, no public modifier). No SDK consumer can reference allHoldouts, init(allholdouts:), or any removed member.
2. Backward-compatible datafile parsing — localHoldouts is decoded with decodeIfPresent defaulting to []. Old datafiles without the key parse identically to before; every holdout lands in the global section, matching pre-change behavior.
3. No removed production code paths — allHoldouts and applyFlatList were only used in tests; zero production callers existed. The only production init path (ProjectConfig.updateProjectDependentProps) already called init(globalHoldouts:localHoldouts:).
4. Holdout evaluation logic unchanged — DefaultDecisionService reads holdoutConfig.getGlobalHoldouts() and holdoutConfig.getHoldoutsForRule(ruleId:), both of which are preserved with identical semantics.
5. Global holdout stripping is defensive — if a backend bug places includedRules on a global-section entry, the SDK now logs a warning and strips it (rather than silently broadening scope). This makes misconfiguration visible without changing behavior.

Changes

• Sources/Data Model/Project.swift — decode the new top-level localHoldouts array; default to [] when absent. Include it in CodingKeys and Equatable.
• Sources/Data Model/HoldoutConfig.swift — section-aware mapping. New init(globalHoldouts:localHoldouts:). Global-section entries get includedRules stripped (with warning if non-nil); local-section entries without includedRules are logged and excluded. Removed allHoldouts property and init(allholdouts:) (no production callers).
• Sources/Data Model/ProjectConfig.swift — construct HoldoutConfig from both project.holdouts and project.localHoldouts sections.
• Sources/Data Model/Holdout.swift — docstring on isGlobal clarifies that section membership is the source of truth.

Tests

• Tests/OptimizelyTests-DataModel/HoldoutConfigTests.swift — new tests for the section-aware semantics: global-section includedRules are stripped/ignored, local section without includedRules is excluded, mixed valid/invalid local entries, both-sections partition enforced, backward compat with no localHoldouts section.
• Tests/OptimizelyTests-DataModel/ProjectTests.swift — new tests for localHoldouts decoding (missing key defaults to []; present key decodes alongside holdouts).
• Tests/OptimizelyTests-DataModel/ProjectConfigTests.swift — updated existing testGenerateProjectConfigMapWithHoldouts to put local holdouts in the new localHoldouts section.
• All holdout test files migrated from allHoldouts setter to HoldoutConfig(globalHoldouts:localHoldouts:).
• Removed redundant project.holdouts dual-write pattern in tests.
• Fixed didSet ordering issues where project.sendFlagDecisions mutation wiped holdoutConfig.

Reference PRs

• Python: [AI-FSSDK] [FSSDK-12760] Add localHoldouts datafile section for backward compat python-sdk#517
• Ruby: [AI-FSSDK] [FSSDK-12760] Add localHoldouts datafile section ruby-sdk#400

Jira

FSSDK-12760

[muzahidul-opti]

Looks good to me. I made clean up on top of the AI workflow:

• Removed allHoldouts property and init(allholdouts:) — no production code used them; only tests. The sole production path (ProjectConfig.updateProjectDependentProps) already used init(globalHoldouts:localHoldouts:).

• Removed applyFlatList() — only existed to support allHoldouts. With it gone, all holdout parsing goes through applySections() directly.

• Cleaned up comments — removed verbose boilerplate doc comments and replaced JIRA ticket references (FSSDK-12760) with generic descriptions.

• Added warning log when a global-section entry has includedRules set (before stripping it), making datafile misconfiguration visible.

• Migrated all ~60 test sites from allHoldouts setter to HoldoutConfig(globalHoldouts:localHoldouts:).

• Removed redundant project.holdouts dual-write in tests — the didSet rebuild was immediately overwritten by the explicit holdoutConfig assignment.

• Fixed didSet ordering bugs where project.sendFlagDecisions mutations wiped the test's holdoutConfig.

• Why it's safe: HoldoutConfig is internal (not public API), zero production callers used the removed members, datafile parsing is backward compatible, and holdout evaluation logic is unchanged.