Replace SemaphoreSlim with Parallel.ForEachAsync, rename MaxConcurrentPageFetches

[donaldgray]

Summary

• Replaces the manual SemaphoreSlim + Task.WhenAll concurrency pattern in TextBuildJob.ProcessPagesAsync with Parallel.ForEachAsync
• Consolidates three per-format fetch methods (FetchXmlWithSemaphoreAsync, FetchVttWithSemaphoreAsync, FetchAnnotationPageWithSemaphoreAsync) into a single FetchPageAsync
• Renames MaxConcurrentAltoFetches → MaxConcurrentPageFetches since the limit applies to all text formats (ALTO, VTT, AnnotationPage), not just ALTO
• Adds a Builder API configuration reference table to README.md

Why Parallel.ForEachAsync over SemaphoreSlim

Less boilerplate — the concurrency limit is a single ParallelOptions property rather than a SemaphoreSlim that must be created, passed into every method, and released in a finally block.

Lazy task scheduling — Parallel.ForEachAsync only starts a new task when a slot is free, so it never allocates more than MaxDegreeOfParallelism concurrent tasks at once. The previous approach created a Task object for every page upfront (all queued on the semaphore), which wastes memory on large manifests.

Clearer intent — "process this collection with bounded parallelism" is exactly what Parallel.ForEachAsync was designed to express. A semaphore is a lower-level primitive that requires the reader to mentally reconstruct that intent.

What is unchanged

• MaxConcurrentPageFetches default remains 8; behaviour is identical
• Per-page errors are still caught inside the delegate and accumulated without aborting the job
• Cancellation still propagates via ShutdownToken
• Results are written into a pre-allocated ordered array so TextBuilder still receives canvases in original canvas sequence

🤖 Generated with Claude Code