feat(gemini): add service_tier support for Flex Inference

[smittie2000]

Description

Add serviceTier provider option to the Gemini provider's Text, Stream, and Structured handlers, enabling Google's Flex Inference API.

Flex Inference provides 50% cost reduction for latency-tolerant workloads (batch processing, background summarization, etc.) by opting into Google's flex service tier.

Usage

Prism::text()
    ->using(Provider::Gemini, 'gemini-2.5-flash')
    ->withPrompt('Summarize this document...')
    ->withProviderOptions(['serviceTier' => 'flex'])
    ->asText();

Note: The Gemini API requires lowercase 'flex' — uppercase 'FLEX' is rejected despite appearing in some Google documentation examples. Also, Flex Inference has a latency target of 1-15 minutes, so users should increase their timeout accordingly.

Implementation

• Follows the same pattern as the existing OpenAI service_tier implementation
• Passes service_tier at the top level of the Gemini REST request body (not inside generationConfig)
• Uses Arr::whereNotNull — when serviceTier is not set, the key is omitted entirely (zero impact on existing requests)
• Error handling already covered: Gemini.php handles 429 (PrismRateLimitedException) and 503 (PrismProviderOverloadedException)

Provider option key naming

Uses serviceTier (camelCase) to follow the Gemini provider's existing convention (thinkingConfig, thinkingBudget, cachedContentName, safetySettings).

Related

• Related: Feature request: Allow custom provider request parameters #925 (custom provider request parameters — this is a concrete implementation of that pattern for one specific key)
• Prior art: [OpenAI] - Allow Service Tier to be set in Provider Options #689 (OpenAI service_tier support)

Breaking Changes

None