[Tsavorite] Handle ObjectAllocator device read failures

[Ted Hart (TedHartMS)]

Root Cause

ObjectAllocatorImpl.AsyncReadPageCallback logged device read failures but signaled completion without preserving the error. Its header-read and partial-sector read-back callers therefore continued after failed reads. Zero-error short reads were also treated as successful completions.

Description of Change

Device read completion now carries the error code, transferred byte count, and exception, and existing-data reads succeed only when the requested length is transferred.

Key changes:

• Convert zero-error short reads into EndOfStreamException failures
• Throw before main-log or object-log truncation when the page-header read fails, preventing success-only OnTruncate callbacks and allowing a later retry
• Route partial-sector flush read failures through the existing flush failure pipeline without submitting a replacement write or advancing FlushedUntilAddress
• Keep front-partial snapshot writes safe by using the zero-filled pooled buffer for the unwritten prefix of a fresh snapshot device
• Include the caught exception's full ToString() output in the OnPagesClosedWorker fatal message so CI logs contain its type, message, inner exceptions, and stack trace even when the runtime omits the FailFast exception argument; formatting runs inside try/finally so a broken custom ToString() cannot bypass fail-fast termination
• Add deterministic numeric-error and short-read tests for flush, truncation, retry, and snapshot behavior

Key Technical Details

Affected types:

• PageAsyncReadResult<TContext> - carries the full device completion result
• ObjectAllocatorImpl<TStoreFunctions> - validates exact reads and prevents unsafe follow-on work
• AllocatorBase<TStoreFunctions, TAllocator> - preserves complete page-close exception diagnostics in fail-fast output
• ObjectAllocatorDeviceFailureTests - injects device errors and EOF-style short reads

Testing

• Regression fixture passes on .NET 10 and .NET 8
• Related object-log deletion, scan, compaction, and snapshot recovery tests pass on .NET 10
• Tsavorite.core builds successfully for .NET 8 and .NET 10
• Changed-file formatting verification passes

Issues Fixed

Fixes #2073

[Copilot]

Pull request overview

Adds reliable ObjectAllocator device-read failure propagation to prevent truncation and flush corruption.

Changes:

• Propagates device errors, short reads, and exceptions.
• Stops dependent flush/truncation operations after failed reads.
• Adds regression tests for failure, retry, and snapshot paths.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
ObjectAllocatorDeviceFailureTests.cs	Adds deterministic device-failure regression tests.
PageAsyncResultTypes.cs	Carries complete read completion details.
ObjectAllocatorImpl.cs	Validates reads and safely routes failures.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.