fix(numeric): replace decompose_float with stdlib formatting in format_float

[Chronolapse411]

Summary

Fix data corruption in float-to-text conversion by replacing the custom decompose_float digit-extraction algorithm with Rust's standard library formatting.

Fixes #6365

Problem

format_float() in core/numeric/mod.rs uses a custom decompose_float algorithm based on DoubleDouble arithmetic for extracting decimal digits from floating-point numbers. This algorithm produces different last-digit rounding than SQLite's internal sqlite3FpDecode, which uses long double / 80-bit extended precision.

When Turso formats a float (e.g., during text affinity conversion via CAST(x AS TEXT)), the resulting string can differ from what SQLite produces:

Input	Turso (before)	SQLite
-8487739174.3030205	-8487739174.3030205	-8487739174.30302
9.93e-322	9.93071948140905e-322	9.93071948140906e-322

When these differently-formatted strings are re-parsed into floats, the resulting bit patterns differ (up to 7 ULPs), causing differential test failures and real data corruption.

Fix

Replace decompose_float-based digit extraction with Rust's format!("{:.14e}"), which uses industry-standard Grisu3/Dragon4 algorithms. These produce the same digit sequences as SQLite's sqlite3FpDecode and C's printf("%.15g").

The new format_float implementation:

1. Uses format!("{:.14e}", v) to get 15 significant digits
2. Applies %g-style fixed vs. scientific notation switch (exponent range [-4, 14])
3. Strips trailing zeros while preserving .0 for integer-valued floats
4. Handles NaN, ±Inf, and 0.0 edge cases

Also removes the now-unused FloatParts enum and decompose_float function (dead code cleanup).

Verification

• ✅ 1504 turso_core unit tests pass (0 failures)
• ✅ clippy passes with -D warnings (0 warnings)
• ✅ 3 new float formatting tests added (test_decode_float, test_format_float_matches_sqlite, test_format_float_edge_cases)
• ✅ Differential simulator seed 6017 (previously failing): simulation succeeded
• ✅ Seed sweep 6010–6025: 0 differential mismatches (6 PASS, 10 timeout-only)

[turso-bot]

Please review @pereman2

[github-actions]

Fossier: PR Auto-Closed

This PR was automatically closed because @chronolapse411 did not meet the trust threshold for this repository.

Score Breakdown

Total Score: 36.4/100 | Confidence: 100% | Outcome: DENY

Signal	Value	Score	Weight
account_age	323	0.89	0.07
public_repos	28	1.00	0.04
contribution_history	28	0.14	0.04
follower_ratio	0.0	0.00	0.04
bot_signals	False	0.50	0.03
open_prs_elsewhere	17	0.00	0.07
closed_prs_elsewhere	13	0.00	0.08
merged_prs_elsewhere	7	1.00	0.06
prior_interaction	0	0.00	0.24
activity_velocity	3	0.50	0.06
pr_content	...	1.00	0.06
commit_email	no_email	0.50	0.03
pr_description	...	0.40	0.04
repo_stars	18180	0.30	0.03
org_membership	0	0.20	0.02
commit_verification	...	1.00	0.03
contributor_stars	0	0.00	0.03

Appeal

If you believe this is a mistake, please open an issue to request manual review. A maintainer can vouch for you by adding your username to the VOUCHED.td file.

You can also reach the maintainers at: https://discord.gg/mNUkCHs2R to appeal this decision

[codspeed-hq]

Merging this PR will degrade performance by 48.47%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 3 improved benchmarks
❌ 8 regressed benchmarks
✅ 300 untouched benchmarks
⏩ 105 skipped benchmarks¹

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	format_float_infinity	1,172.8 ns	984.4 ns	+19.13%
❌	Simulation	format_float_integer	5.3 µs	10.3 µs	-48.47%
⚡	Simulation	format_float_neg_infinity	1,201.1 ns	954.4 ns	+25.84%
❌	Simulation	format_float_negative	5.4 µs	7.6 µs	-29.55%
❌	Simulation	format_float_precision_edge	5.2 µs	7.6 µs	-31.94%
❌	Simulation	format_float_very_small	6.3 µs	9.1 µs	-30.7%
❌	Simulation	format_float_simple	5.3 µs	7.6 µs	-29.67%
❌	Simulation	format_float_scientific_needed	7.1 µs	9.1 µs	-21.97%
⚡	Simulation	format_float_zero	1,107.2 ns	859.2 ns	+28.87%
❌	Simulation	format_float_very_large	6.2 µs	9.2 µs	-32.54%
❌	Simulation	length_float	7.2 µs	9 µs	-20.46%

---

_{Comparing Chronolapse411:fix/format-float-precision (b71b0ba) with main (9b3dc16)}

Footnotes

1. 105 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩