v0.0.5

This release fixes a class of silent-failure bugs across the query tools where filter and enum arguments documented in the spec either did nothing or fell through to a default without any error.
Every gap closes with a structured ToolError that echoes the offending value and lists the accepted set, so the LLM can correct in one retry instead of guessing why a query came back empty.

Fixed

• (query) wire process filter (name + sub-pid) (#43).
  ProcessFilter::Name was a TODO that always returned false, and pid: parsed only bare u64, so samply's .N sub-pid form fell through to the broken name path.
  Every process= value silently returned 0 samples.
  Name now matches each thread's processName; Pid widens to raw::Pid so a bare pid matches every sub-process sharing the OS pid and a suffixed pid pins to one.
  Unknown values now surface process_not_found with available_processes instead of an empty result.
• (query) wire time_range filter through sample iteration (#48).
  time_range was parsed into Filter and documented in the spec for top_functions, call_tree, stacks_containing, folded_stacks, and compare_profiles, but no aggregator consulted it — the unsliced full profile came back regardless.
  Per-sample gating now lives on Profile::stack_indices and uses samples.absolute_times() / markers.start_time to drop items outside the inclusive range.
  Fully-disjoint or inverted ranges return out_of_bounds via the new validate_time_range; partial overlap clamps silently per the design spec.
• (query) reject unknown sort_by/group_by/align_by values (#49).
  Five tool entry points used catch-all _ => arms that silently fell back to the default on unrecognized string-enum values — a sort_by="selfTime" typo silently ranked by self-time anyway, group_by="lib" silently grouped by function.
  All five sites now route through a parse_string_enum helper that returns ToolError::InvalidValue { field, value, accepted } with the full accepted set.
• (query) reject malformed pid:/tid: prefix instead of name fallback (#50).
  tid: and pid: opt into integer matching, but the parser silently fell back to a literal name match when the rest didn't parse — process="pid:abc" then surfaced process_not_found listing every real process, none of which resembled the input.
  parse_filter is now fallible: malformed prefixes return invalid_value with the accepted syntax (tid:NNN, pid:NNN, pid:NNN.M, <thread/process name>).
  Bare-name matches without a prefix still work unchanged.

Other

• enable clippy::match_wildcard_for_single_variants as defensive hygiene against future enum exhaustiveness mistakes
• drop redundant #![allow(dead_code)] from src/profile/event_source.rs and src/query/event.rs after the iterator refactor