fix(pixel-chat-context): coerce float and string token limits in context trimmer

[vaibhavsrv]

Why this matters

In ods/extensions/services/dashboard-api/pixel_chat_context.py, the public models ContextUsage and ContextModel inherit from _Projection, which declares strict=True in its Pydantic configuration. Because of strict mode, integer token fields such as used, window, and contextWindow strictly reject float values. When upstream inference runtimes (e.g. llama-server or vLLM) or JSON parsers serialize token statistics as decimal numbers (such as 100.0 or 2048.0), Pydantic throws an unhandled ValidationError: Input should be a valid integer [type=int_type, input_type=float], causing persistent context queries to fail with HTTP 500 errors.

This change adds @field_validator(..., mode="before") hooks to ContextUsage and ContextModel. When incoming numeric values are floats representing whole integers (value.is_integer()), they are automatically coerced to int prior to validation. Non-integer floating point numbers continue to be strictly rejected.

Validation

• Tested baseline reproduction against unpatched code: passing {"used": 100.0, "window": 2048.0} in context usage triggered a Pydantic ValidationError.
• Tested post-fix behavior: exact integer floats are coerced to int and validated successfully; non-integer floats (e.g. 100.5) remain rejected.
• Telemetry statement: "Dashboard-api suites: 1 passed. New-test Ruff, py_compile, and diff checks pass; new regression wired into Linux CI."

Overlap check

• fix(model-contract): coerce integer float token limits in projection #5606 (open): Coerced integer float token limits in model_contract.py.
• fix(pixel-settings): reject active token limits exceeding provider capacity #5696 (merged): Clamped active token limits in pixel_settings/contract.py.
• fix(dashboard-api): tolerate evicted or missing attempts during pixel chat cancel #5695 (merged): Handled missing/evicted attempts in chat cancellation.
• fix(dashboard-api): ensure trailing buffered chunk is newline-terminated before done marker #5946 (open): Ensured trailing buffered chunk newline termination in routers/pixel.py.
  This PR targets Pydantic context projection in pixel_chat_context.py and does not conflict with adjacent PRs.

Risk / AI disclosure

AI-assisted defect analysis, implementation, and unit regression testing. This standardizes numerical type coercion for whole-number floats and does not alter persistence schemas, model endpoints, or container routing. Independent human review remains a gate.

Follow-up integration evidence

Composed cleanly on top of #5569, #5871, and #5875 at HEAD without conflicts. Production and test diffs passed together; adjacent pixel-agent and pixel-settings checks remain intact.
Backlog composition was local-only (production/test diffs, excluding workflow/Makefile wiring); it is not an upstream merge or independent human approval. Declared live-review gates remain open.