Fix #1776: local pointer in non-default Writer string scan loops

[pjfanning]

Fixes #1776. Builds on #1682.

Change

Applies #1682's _writeString2 fix to the two non-default WriterBasedJsonGenerator string scan loops: scan with a local pointer and write _outputTail back once, instead of ++_outputTail per char. C2 doesn't keep the field in a register across loop iterations, so the old loops stored it on every char.

• _writeStringASCII(int, int): used with JsonWriteFeature.ESCAPE_NON_ASCII / highestNonEscapedChar(...)
• _writeStringCustom(int): used with custom CharacterEscapes. The getEscapeSequence(c) call inside the loop has no access to generator state, so a local pointer is safe there too; no write-back is needed before the call (contrary to what I suggested in Use local pointer in non-default WriterBasedJsonGenerator string scan loops #1776).

The _writeSegment* and char[] counterparts already use local pointers.

Options

This branch is built on #1682 (head 7a7a330b3), so until #1682 is merged the diff here also shows #1682's commits. The only new commit is 4284dad. Two ways to handle it:

1. Fold into Fix #1680: drop per-character escape-table load in Writer string loops #1682: cherry-pick 4284dad onto Fix #1680: drop per-character escape-table load in Writer string loops #1682's branch and close this PR. It is the same pattern and the same file, so it can be reviewed together.
2. Hold back: merge Fix #1680: drop per-character escape-table load in Writer string loops #1682 first, then merge this later (e.g. 3.4.x, as Use local pointer in non-default WriterBasedJsonGenerator string scan loops #1776 suggests). These are non-default configurations, so there's no urgency. Once Fix #1680: drop per-character escape-table load in Writer string loops #1682 is merged this PR's diff shrinks to the one commit.

Either is fine with me.

Testing

• Generator, escaping, write, JSONP and non-ASCII test classes: 402 tests pass.
• Also ran Fix #1680: scan + bulk-copy fast path for longer String values in UTF8JsonGenerator #1775's StringBulkCopy1680Test against this branch. It compares Writer output with UTF8JsonGenerator (unchanged here) under ESCAPE_NON_ASCII, highestNonEscapedChar(0xFF) and JsonpCharacterEscapes, which are exactly the changed paths. Both tests pass.

JMH (JDK 17, i7-8550U)

writeString(String) x 200 ASCII values per op into a no-op Writer, us/op, #1682 vs this PR:

configuration	len	#1682	this PR	change
JsonpCharacterEscapes	100	41.9 ± 1.8	29.8 ± 5.7	-29%
JsonpCharacterEscapes	1000	261.7 ± 20.1	211.7 ± 20.0	-19%
ESCAPE_NON_ASCII (3 forks)	100	44.3 ± 1.9	40.5 ± 2.1	-8%
ESCAPE_NON_ASCII (3 forks)	1000	252.6 ± 11.8	227.0 ± 14.9	-10%

No release-notes entry yet: it depends on which option is chosen and which version it lands in.

🤖 Generated with Claude Code

[github-actions]

📈 Overall Code Coverage

Metric	Coverage	Change
Instructions	86.49%	📉 -0.040%
Branches	80.49%	📉 -0.130%

Overall project coverage from JaCoCo test results. Change values compare against the latest base branch build.

[pjfanning]

Folded into #1682 (commit 4284dad, results: #1682 (comment)), so closing this one.