Expose MaxCharacterCount as a WolframLanguageEvaluator tool option

[basnijholt]

Problem

WolframLanguageEvaluator hard-codes "MaxCharacterCount" -> 10000 at both WolframLanguageToolEvaluate call sites, so every tool result is truncated at ~10 kB with no way to configure it. (There's an adjacent FIXME: Expose this as an option in the file.)

This particularly hurts image/plot workflows in headless MCP deployments: a rasterized plot returned as base64 blows through 10 kB immediately, forcing agents into awkward workarounds (chunking base64 across many evaluations and reassembling client-side). Most modern agent harnesses already handle large tool outputs themselves (spilling to files, summarizing, or capping in the client), so a hard server-side 10k limit mostly gets in the way.

Change

Wire MaxCharacterCount through the existing tool-options mechanism, following the same pattern as Method / TimeConstraint / MaxSessionCount:

• $maxCharacterCount := toolOptionValue[ "WolframLanguageEvaluator", "MaxCharacterCount" ]
• "MaxCharacterCount" -> 10000 added to $defaultToolOptions[ "WolframLanguageEvaluator" ] — default behavior is unchanged
• both call sites now pass $maxCharacterCount

Users can then raise (or lower) the cap via the standard "ToolOptions" plumbing, e.g.:

InstallMCPServer[
    "ClaudeCode",
    "WolframLanguage",
    "ToolOptions" -> <| "WolframLanguageEvaluator" -> <| "MaxCharacterCount" -> 100000 |> |>
]

Tests

• Extended DefaultToolOptions-WolframLanguageEvaluator with the new default.
• Added ToolOptionValue-FallbackToDefault-MaxCharacterCount and ToolOptionValue-UserOverride-MaxCharacterCount, following the existing Method/TimeConstraint tests. (TestID location suffixes left to Scripts/FormatFiles.wls per docs/testing.md.)

Manually verified against a licensed Wolfram Engine via PacletDirectoryLoad: default unchanged (10000); with $toolOptions overriding to 50000, StringRepeat["a", 60000] returns ~50k characters instead of truncating at 10k.