[mrizvi96]

Fixes #206.

Changes

• onair/data_handling/parser_util.py
  • floatify_input(): catches (ValueError, TypeError) around float()
    (no more TypeError crash on None/dict/list); non-finite results
    ("nan"/"inf" strings, NaN/Inf floats) never enter the output;
    unparseable values — including the "-" stale-data sentinel — are
    substituted with math.nan plus a print_msg WARNING naming the value
    (or dropped entirely when remove_str=True). The silent 0.0 fallback
    is removed everywhere; policy documented in the docstring.
  • convert_str_to_timestamp(): rejects non-str inputs up front;
    MM:SS fallback requires exactly two colon-separated parts, both
    integer-parseable, minute and second in 0..59 — else ValueError.
    "12:34:56" is no longer silently truncated; no IndexError /
    AttributeError leaks.
• plugins/kalman/kalman_plugin.py — Plugin.update(): frames whose length
  does not match len(self.headers) are rejected with a WARNING naming both
  lengths; non-finite values are skipped per-attribute, so windows contain
  only finite floats (a skipped sample advances no window).
• No public signatures changed; stdlib math only; no new dependencies.
• Legacy tests updated ONLY where they encoded the removed 0.0-masking or
  mock-only behavior (every edit marked with an inline # updated: comment;
  no assertions weakened).

Validation

All 371 existing tests pass on the unchanged baseline (python -m pytest test/ -q); 401 pass with this change — 30 new tests:

• 10 parser properties (never-raises on JSON-grade scalars, length
  preservation, finite-float identity, "-" sentinel policy, NaN/Inf
  normalization, timestamp validity), each at max_examples=1000, deadline=None;
• 4 kalman property tests: a RuleBasedStateMachine feeding
  valid/junk/short/long/NaN-spiked frames (invariants: update() never
  raises, windows stay finite and length-aligned) plus 3 frame-rejection
  properties;
• 16 example-based regression tests (8 per module) pinning exact epochs,
  exact warning text, and the frame-length guard.

Mutation testing (mutmut), before → after, two runner configurations
(legacy files only; legacy + property files derandomized), each run twice
with identical results:

Scope	Before	After
parser_util.py (52 mutants)	1 survivor in floatify_input	0 (hardened code: 71/71 killed)
Plugin.update() in-scope mutants	2 survivors	0

Full evidence — survivor catalogs, pre-patch falsifying examples, kill
tables, policy semantics, legacy-test change log — included in this PR
(doc/telemetry-hardening-report.md).

Notes for reviewers

• Draft PR to solicit feedback, per contributing guidance for larger
  changes; happy to split into parser / kalman PRs if preferred.
• The MM:SS epoch uses naive datetime.timestamp() (host-local time).
  This is pre-existing upstream behavior, intentionally preserved; flagged
  in the report (§11.3) as a candidate decision for maintainers.
• Individual CLA per the contributing instructions: completed/submitted.