fix: error rows written to wrong file and default sensors mismatch

[k-rister]

Summary

• Fix error rows written to wrong file path: error rows used ${device//://} (colon→slash, creating nested dirs) while normal rows used ${device//:} (remove colons). Data went to different files.
• Align default sensors with documentation: code defaulted to 1,2,6,127 but README and all usage examples document 1,2,6. Sensor 127 is not a valid MVCR sensor index.

Note: the Jira ticket also flagged the CDM metric class as potentially wrong (count vs power), but investigation shows count is correct per the CDM specification — power watt readings are a gauge/level measurement which falls under count. tool-power using power may itself be a bug.

Closes #4

Test plan

• CI passes
• Verify error and normal CSV rows go to the same file
• Verify default sensor collection matches 1,2,6

🤖 Generated with Claude Code

[k-rister]

Updated per review feedback:

1. Sensor 127: Restored — the README was out of date, not the code. The sensor 127 commit is dropped entirely (no net change to mlxreg-start).

2. File path separator: Adopted Hugh's underscore suggestion (${device//:/_}). Produces 0000_03_00.0.csv — unique and reversible. Applied to all four locations in mlxreg-collect (header creation, normal rows, both error paths). Updated the post-process filename parser to use the simpler replace('_', ':') instead of the regex approach.

Force-pushed a single clean commit.

[atheurer]

File path fix looks good — colon→underscore is the right choice and the post-process reverse mapping is cleaner.

One note: the PR description mentions aligning default sensors from 1,2,6,127 to 1,2,6, but I don't see that change in the diff. Was it dropped or is it planned for a separate PR?

[k-rister]

One note: the PR description mentions aligning default sensors from 1,2,6,127 to 1,2,6, but I don't see that change in the diff. Was it dropped or is it planned for a separate PR?

Apparently 127 is a valid sensor and the documentation was lacking so it remains.