fix(cli): tool cursors cannot be redeemed by a later tool call

[ScriptedAlchemy]

Problem

A next_cursor printed by tracedecay tool cannot be redeemed by the next tracedecay tool invocation. The follow-up call returns a typed unavailable problem instead of the next page, so a CLI user can never read past the first 10 rows of a relation page. The same cursor works inside one tracedecay serve MCP session.

Reproduction

Shipped CLI built from master at f60e18f plus #2130's branch (the branch does not touch cursors). Isolated profile (HOME, TRACEDECAY_DATA_DIR, TRACEDECAY_GLOBAL_DB, TRACEDECAY_DAEMON_SOCKET in a scratch dir), one daemon, starship@cc825b00 as the corpus:

tracedecay tool tracedecay_callees --args '{"node_id":"<src/modules/mod.rs::handle id>","maximum_depth":1,"meta":{"projection":"evidence","order":"source_position"}}' --json
# payload.total = 112, returned = 10, next_cursor = ccq2.…
tracedecay tool tracedecay_callees --args '{"node_id":"<same>","maximum_depth":1,"meta":{"projection":"evidence","order":"source_position","cursor":"<that next_cursor>"}}' --json

Observed on the second call (retried after 1 s, same result):

"problem":{"kind":"unavailable","code":"application.surface.unavailable","message":"The application service for this operation is unavailable",...,"unavailable_classification":"authority","retryable":true,"retry":"after_delay"}

Expected: the second page (rows 11 through 20). Sending the same two requests as tools/call over one tracedecay serve -p . stdio session pages through all 112 rows.