refactor: add msat model to encapsulate conversions

[ovitrif]

This PR introduces an MSat inline value class that replaces all scattered (x + 999u) / 1000u ceiling and x / 1000u floor patterns with self-documenting conversions.

Description

The msat-to-sat conversions were spread across 13 files with no domain type to explain what's happening. After PR #879 added more ceiling conversions, it became harder to understand the intent behind (x + 999u) / 1000u without reading the PR context. The new MSat value class (zero runtime overhead) makes the rounding intent self-documenting.

• Adds MSat inline value class in models/ with ceil() and floor() functions
• Adds msatCeilOf() and msatFloorOf() top-level sugar functions for concise call sites
• Replaces all ceiling patterns ((x + 999u) / 1000u, msatsToSatsCeil) with msatCeilOf(x)
• Replaces all floor patterns (x / 1000u, x / MSATS_PER_SAT) with msatFloorOf(x)
• Removes msatsToSatsCeil() function and MSATS_PER_SAT constant from Lnurl.kt
• Adds MSat.PER_SAT constant for the sat→msat direction
• Adds MSatTest.kt with 10 test cases

Preview

N/A — no UI changes, pure refactor.

QA Notes

All tests listed and checked below were verified manually, this is a refactoring PR which shouldn't require extensive retesting by reviewer.

Tests

All tests should use msat values:

• LNURL-Pay (QuickPay ON & OFF) using fix: preserve lnurl msat precision bitkit-docker#5
• LNURL-Withdraw (QuickPay ON & OFF) using fix: preserve lnurl msat precision bitkit-docker#5
• Bolt11 (QuickPay ON & OFF) using feat: support msat amounts in bolt11 generate endpoint bitkit-docker#6
• regression: Bolt11 zero-amount invoice

[claude]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

[jvsena42]

@claude Give me a suggestions of instruction to CLAUDE.md to always use the convention from this class in sats operations

[claude]

Claude finished @jvsena42's task in 52s —— View job

---

CLAUDE.md instruction suggestions for MSat convention

Here are the suggested rules to add to the Rules section in CLAUDE.md, following the existing style:

- ALWAYS use `MSat(value).ceil()` or `msatCeilOf(value)` when converting millisats to sats rounding **up** (payment amounts, display amounts, minimum bounds); NEVER use raw `(x + 999u) / 1000u`
- ALWAYS use `MSat(value).floor()` or `msatFloorOf(value)` when converting millisats to sats rounding **down** (fees, upper bounds, capacity); NEVER use raw `x / 1000u` or `x / MSATS_PER_SAT`
- ALWAYS use `MSat.PER_SAT` when multiplying sats to millisats; NEVER use the raw literal `1000u` for sat→msat conversions
- NEVER mix `.ceil()` and `.floor()` for the same semantic value — use `.ceil()` for minimum/payment amounts and `.floor()` for maximum/fee amounts

Rationale per rule:

Rule	Why
msatCeilOf for payment amounts	Rounding down a minimum produces a sats value that, when multiplied back × 1000, falls below the protocol minimum and can cause LNURL-Withdraw rejections
msatFloorOf for fees/upper bounds	Rounding up a fee overestimates cost; rounding up a channel capacity could exceed actual balance
MSat.PER_SAT for sat→msat	Prevents magic number 1000u from reappearing after MSATS_PER_SAT was removed in this PR
No mixing ceil/floor	A wrong rounding direction is a silent correctness bug — the rule forces the author to consciously pick the right one