fix(agentspan): map handoff toolType to INLINE instead of silent SIMPLE default

[kowser-orkes]

Pull Request type

• Bugfix
• Feature
• Refactoring (no functional changes, no api changes)
• Build related changes
• WHOSUSING.md
• Other (please describe):

NOTE: Please remember to run ./gradlew spotlessApply to fix any format violations.

Changes in this PR

Bug: ToolCompiler.TYPE_MAP has no entry for toolType="handoff" (the SWARM/hybrid transfer
tool's type since PR #1356). compileToolSpecs silently falls through to its "SIMPLE" default,
so every {source}_transfer_to_{peer} tool's LLM-facing spec says "type": "SIMPLE" — directly
contradicting the "deliberately not a worker" comment on the same .toolType("handoff") call.

Impact: cosmetic to the server's own dispatch (transfer calls are intercepted by the INLINE
checkTransferTask/toolRouter before dynamic-fork dispatch ever runs), but it's the field
client SDKs read to decide "does this tool need a local worker" — and it's wrong. One half of the
fix for #1363 (stateful SWARM handoffs hanging forever); paired python-sdk PR fixes the client
side that this bug misleads.

Fix:

• TYPE_MAP: add "handoff" → "INLINE", matching how handoff_check/check_transfer are
  already typed.
• compileToolSpecs: WARN-log unrecognized toolTypes instead of silently defaulting to
  "SIMPLE" — a silent default to the one type that implies "needs a worker" is exactly what let
  this regress unnoticed for 9+ days.
• Tests: new testCompileToolSpecs_Handoff (direct coverage) and
  testCompileToolSpecs_compilerOwnedToolTypesHaveExplicitMappings (pins all 3 compiler-owned
  toolTypes so a future unmapped one is caught here). Fixed one pre-existing
  SwarmHandoffCompilerContractTest assertion that had hardcoded this bug ("SIMPLE") as expected
  behavior.

Not fixed by this PR alone: requiredWorkers (computed from
WorkflowDef#collectSimpleTaskNames(), which reads the compiled graph task's type, not the LLM
tool spec) already excluded these names, confirmed live both before and after this change. This PR
only fixes the LLM-facing spec field.

Issue #1363

Alternatives considered

• Restore server-side TaskDef registration for the transfer tool — rejected: the transfer task
  is never dynamically forked/scheduled in the first place (checkTransferTask/toolRouter
  interception), so a registered TaskDef would just be dead clutter.
• Introduce a new task-type marker instead of reusing "INLINE" — rejected: INLINE is already
  what handoff_check/check_transfer use for "resolved server-side, no worker."

Removing