fix: send the action ids a protocol maps to names

[u9g]

On 26.1 client_command.actionId and use_entity.hand are mapper fields:

"packet_client_command": ["container", [{"name": "actionId", "type": ["mapper", {"type": "varint", "mappings": {"0": "perform_respawn", "1": "request_stats", "2": "request_gamerule_values"}}]}]]
"packet_use_entity":     ["container", [{"name": "target", ...}, {"name": "hand", "type": ["mapper", {"type": "varint", "mappings": {"0": "main_hand", "1": "off_hand"}}]}, ...]]

Writing the numeric id there throws inside the serializer:

Error: SizeOf error for undefined : 0 is not in the mappings value
    at Object.packet_use_entity ...

The packet is dropped on the floor, so on 26.1 the bot cannot respawn (bot.respawn(), the death handler, the win_game handler), cannot request stats, and bot.activateEntity() does nothing at all — right-clicking an NPC on a 26.1 server looked like the server ignoring us. entity_action has been string-mapped since 1.21.6 and physics.js already handles it, but bed.js's wake() still wrote actionId: 2, which both fails to serialize there and means stop_sprinting under the new mapping.

game.js also wrote { action: 0 }, a field name that does not exist in any protocol version, so the win_game respawn never serialized anywhere.

lib/protocol_ids.js asks the schema whether the field is a mapper rather than going through a feature, so this is right whether minecraft-data keeps 26.1's mappers (PrismarineJS/minecraft-data#1279) or drops them (PrismarineJS/minecraft-data#1278), and it needs no unreleased data. bed.js keeps supportFeature('entityActionUsesStringMapper') to match physics.js, which already uses it for the same packet.

Verified live: with this change bot.activateEntity() on a 26.1 server's NPC produces use_entity {target, hand: 'main_hand', location, sneaking} and the server answers, where before nothing went out.

Tests: three cases in test/internalTest.js covering every tested version. They resolve the packet out of the server's read path before asserting — an assertion thrown inside the packet listener wedges the connection and shows up as an afterEach timeout — and on 1.8-1.11, whose use_entity has no hand field at all, the assertion is just that the packet arrives.

[github-actions]

Tests more than 1.5x slower than master (durations are noisy, so this is informational):

   8648ms ->   15657ms  mineflayer_external 1.10.2v fishing
   2950ms ->   18392ms  mineflayer_external 1.9.4v fishing

[u9g]

#4079 is the same fix. If PrismarineJS/minecraft-data#1278 lands, use_entity.hand needs no feature check, but client_command.actionId (26.1) and entity_action.actionId (1.21.6+) still do.

[u9g]

#4076 (opened first) covers client_command and entity_action with the same schema-introspection approach under a different helper name. What is only here: use_entity.hand in inventory.js, which is the one that actually breaks a bot — on 26.1 bot.activateEntity() serializes nothing at all, so right-clicking an NPC looks like the server ignoring you. If these are consolidated into #4076, that hunk and its test need to come along.