In Ancient Quenya, the nasal [n] assimilated to a following liquid: nl, nr > ll, rr. This change was mentioned in both the Outline of Phonetic Development from the 1940s (OP1) and from the Outline of Phonology from the 1950s (OP2):

Where the original sequence was maintained, after m: l became syllabic (as after stops), and the resultant vowel took its quality from the preceding vowel: amla > amala; imla > imila etc. But nl became ll; and in this case n-l is only a later form where it was felt desirable to restore the obscured stem. mr where not metathesized probably remained later than ml, and the ultimate product was mar. The phonetic product of nr, however, was rr; the later "restored" form was nar (OP1, PE19/47).

nl > ll; nr > rr; but mr, ml became syllabic and produced mar, mul ... Where n-l, n-r were restored or later produced the result was n-l (with intervening vowel dependent on the vowel before the n) and nar. (OP2, PE19/98).

As noted above, in cases where this sound combination was produced later (e.g. in compounds), the result was a syllabic liquid that produced a vowel. The sound change of nr > rr was rare, however, since frequently nr > rn as part of Quenya metathesis (PE19/47, 97). There are, however, attested examples of nl > ll, such as:

• ᴹ✶nen-le > ᴹQ. nelle “brook” (Ety/NEN).

This sound change can also be seen in some Quenya past tenses produced by nasal-infixion, especially for basic verbs with the consonant l. This was discussed in Tolkien’s notes on Quenya past tense in his description of the Quenya Verbal System from 1948:

[l] usually employs ll (< nl), but ld (< ln) also appears. √TUL “come”: tule, pa.t. tulle, perf. utúlie. √KHAL (cf. halda “high, tall”) in orhale “exalt”, pa.t. orhalde (orhalle), perf. {orhálie >>} orahallie. √OL “grow”, olle “became”, perf. olólie, ólie. Also from weak present: ehtelu- “well, bubble out” (< et­kelu), pa.t. ehtelle, perf. ehtelunelye (see below) or etekélie (PE22/103).

Other past tenses fitting this pattern include:

• ville past tense of ᴹQ. vil- or vil- “to fly” (Ety/WIL).
• tolle “stood” past tense of ᴹQ. tolu- “to stand up” (PE22/114, 117).
• ulle “poured” the (intransitive) strong past tense of ᴹQ. ulya- “to pour” (Ety/ULU; PE22/112).

It seems these ll past tenses were somewhat divorced from their present tense forms, and could be replaced. For example, the more usual past tense of Q. tul- was túle (SD/246, LR/47, PE22/140), probably produced by analogy from the perfect utúlie.

This sound change can also be seen in the assimilated form of the partitive plural suffix -li after nouns ending in -n, such as elelli the partitive plural of elen “star” (PE17/127).

Conceptual Development: There isn’t enough information to determine how this sound combination developed in the Early Quenya in the 1910s and 1920s. However, ll past tenses for verbs with the consonant l were fairly common (appearing beside other past forms with long vowels), which hints that this sound change may already have been established:

• elle “came” past tense of ᴱQ. ele- (MC/215).
• kalle “shone” past tense of ᴱQ. kal- (PE16/143).
• malle past tense of ᴱQ. mala- “am able to” (PE15/67).
• palle “shook” past tense of ᴱQ. pal- (PE16/143).
• pelle past tense of ᴱQ. pele- “surround” (QL/73).
• qalle “died” past tense of ᴱQ. qal- (PE16/143).
• talle past tense of ᴱQ. tala- “carry, bring” (QL/88).
• welle past tense of ᴱQ. wele- “it boils, bubbles” (QL/103).
• yolle past tense of ᴱQ. yolo- “stink” (QL/106).

Thorsten Renk suggested that this Early Qenya past tense pattern was the result of nl > ll in his article on the Quenya Past Tense (QPT).

In Quenya Notes from 1957 (QN: PE17/145) Tolkien considered an alternate development whereby [n] vocalized before [r], much like how primitive velar medial [ŋ] usually became [ɣ] which then vanished with compensatory lengthening ([ŋr] > [ɣr] > [¯r]):

māra “good”, which, as the † byform marna shows, is from manrā, in earliest formations nr > ¯r, (with nasality of preceding vowel subsequently lost) (PE17/162).

Elsewhere mára is given other etymologies, typically from the root √MAG: [magra] > [maɣra] > [māra]. Thus, I think [nr] > [¯r] was a transient idea; there is no evidence of it outside the note above.