feat: canister logs extensions

[mraszyk]

Canister logs extensions:

• New canister setting log_memory_limit bounding the memory used for canister logs: it must be either 0
  or a number between 4096 and 2097152 (2 MiB), inclusively, with the default value 4096.
  The oldest canister logs are purged if the total memory used for canister logs exceeds this value.
  The memory used by the store holding the canister logs is determined by this setting
  (it does not depend on the canister logs actually stored) and counted in the canister's memory usage.
  Hence, raising this setting might require reserving cycles. Changing this setting also resizes
  the store holding the canister logs, which consumes cycles.
  The setting is reset if the canister runs out of cycles.
• canister_status returns the setting log_memory_limit and the memory used by the store holding
  the canister logs in the new field log_memory_store_size of memory_metrics.
• fetch_canister_logs can also be called by canisters via replicated (update) calls.
  It still cannot be called by external users via replicated calls.
• New optional filter argument of fetch_canister_logs restricting the returned logs to a range of
  log indices (by_idx) or timestamps (by_timestamp_nanos).
• The total size of all logs returned by fetch_canister_logs is bounded by an implementation-defined
  constant chosen so as not to exceed the maximum response size (instead of the previous bound of 4KiB).
  If the selected logs do not all fit, an unfiltered read trims the oldest logs (so the response ends
  with the newest log) and a filtered read trims the newest logs (so the response starts with the oldest
  log satisfying the filter).

[pierugo-dfinity]

Claude seems to suggest that we should also update docs/references/management-canister.md

[mraszyk]

Feedback addressed:

• docs/references/ic-interface-spec/management-canister.md: reworded the fetch_canister_logs replicated-call sentence to use "but" instead of "i.e.", and fixed "can be provided and has one of the following variants".
• docs/references/ic-interface-spec/abstract-behavior.md: the "Canister out of cycles" prose now says the allocations and the canister log memory limit are set to zero.
• public/references/ic.did: added the trailing ; after the filter field of fetch_canister_logs_args for consistency with the rest of the file (validated with didc check).
• docs/references/management-canister.md (per review body): added the log_memory_limit row to the canister settings table, added log_memory_store_size to the memory_metrics breakdown, and updated fetch_canister_logs with the new filter parameter, the relaxed caller rules, the response-size bound (replacing the stale 4 KiB cap) and the filtered/unfiltered trimming behavior.

Not addressed: the CanisterId vs Canister_id inconsistency in abstract-behavior.md (pre-existing across the file, left for a separate cleanup).