fix: increase signal in git diff, git log, and json filters (#621)

[pszymkowiak]

Closes #621

Summary

Three filters were cutting signal that LLMs need:

Filter	Before	After
git diff	30 lines/hunk	100 lines/hunk
git log	1 body line	3 body lines
json	Schema only (no values)	Values by default, --schema for types-only

Why

• git diff at 30 lines: LLMs miss the end of refactors and produce incorrect code
• git log at 1 line: BREAKING CHANGE notes, migration instructions lost
• json without values: useless for config debugging (the primary use case)

Test plan

• 979 tests passing, clippy clean
• rtk git diff HEAD~3 — shows full hunks up to 100 lines
• rtk git log -5 — shows 3 body lines per commit
• rtk json — shows values ("rtk", "4.5", etc.)
• rtk json --schema — shows types only (old behavior)
• Tested with phi4:14b (LLM): 8/10 comprehension, correctly reads all values

[pszymkowiak]

LLM Comprehension Test — 57 RTK Commands

Tested all 57 RTK Rust commands with 3 local LLMs to verify filtered output is actionable.

Protocol

For each command: send RTK output to LLM with "You ran <cmd>, output: <output>. <question>" — simulates real AI coding workflow. LLM must answer correctly to PASS.

Results

LLM	Size	Score
llama3.1:8b	8B	57/57 (100%)
phi4:14b	14B	57/57 (100%)
mistral:7b	7B	57/57 (100%)

0% confusion rate — all RTK filtered outputs correctly understood by LLMs as small as 7B. No command blocks the AI from taking the correct next action.

Commands tested

Git (5), Files (8), Data (4), Runners (4), GitHub (3), Network (2), Meta (7), Python (4), Go (2), Docker (2), TOML filters (12), JS/TS (4) = 57 total

Also fixed in this PR

• rtk wget -O - bug (bug: rtk wget interprets -O - flag as URL #716): - was interpreted as URL
• stat TOML filter: strip Device/Birth noise